diff --git a/apps/docs/content/docs/workflows/blocks/agent.mdx b/apps/docs/content/docs/workflows/blocks/agent.mdx
index b880916cd78..dde3a58689c 100644
--- a/apps/docs/content/docs/workflows/blocks/agent.mdx
+++ b/apps/docs/content/docs/workflows/blocks/agent.mdx
@@ -27,6 +27,10 @@ Answer in two sentences, cite the doc you used, and never guess a price.
 
 The model that runs the step. Defaults to `claude-sonnet-4-6`. Type or pick any model from OpenAI, Anthropic, Google, xAI, Groq, Cerebras, DeepSeek, Azure, AWS Bedrock, Google Vertex, or OpenRouter, or a local model through Ollama or VLLM.
 
+For a custom cloud deployment, enter its provider prefix and model ID: `azure/my-deployment`, `azure-anthropic/my-deployment`, `bedrock/my-inference-profile`, or `vertex/my-gemini-model`. The prefix selects the provider and shows its credential fields even when the ID is absent from the catalog. Bedrock accepts full inference profile ARNs after `bedrock/`; Vertex uses the Gemini API and accepts Google model resource names. The deployment must support the selected provider's API. Custom IDs have no catalog pricing or token limits.
+
+Ollama Cloud, OpenRouter, Fireworks, Together AI, Baseten, Ollama, vLLM, and LiteLLM load their available models from the configured provider. New models appear through that discovery without a Sim catalog release. You can also enter a namespaced ID directly, such as `ollama-cloud/deepseek-v4.1-flash`, `openrouter/provider/model`, or `ollama/my-local-model`. Provider prefixes are case-insensitive; the model ID after the prefix keeps its original casing.
+
 ### Files
 
 Files for the model to read: images for a vision-capable model, or documents for text. Upload them on the block, or pass a file from an earlier block, such as an upload trigger or an [API](/workflows/blocks/api) response, with a connection tag.
@@ -110,18 +114,19 @@ Live tool-call chips stream for **OpenAI, Anthropic, Azure Anthropic, Google, Ve
 
 | Provider | Streamed thinking | Models |
 |----------|-------------------|--------|
-| OpenAI | Summaries only — Requires OpenAI organization verification; falls back to no summaries. | `gpt-6-astra`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5-pro`, `gpt-5.5`, `gpt-5.4-pro`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`, `gpt-5.2-pro`, `gpt-5.2`, `gpt-5.1`, `gpt-5-pro`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `o4-mini`, `o3`, `o3-mini`, `o1` |
+| OpenAI | Summaries only — Requires OpenAI organization verification; falls back to no summaries. | `gpt-6-astra`, `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5-pro`, `gpt-5.5`, `gpt-5.4-pro`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`, `gpt-5.3-codex`, `gpt-5.2-pro`, `gpt-5.2`, `gpt-5.1`, `gpt-5-pro`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `o4-mini`, `o3`, `o3-mini`, `o1` |
 | Anthropic | Summaries only — These generations omit full thinking; Sim requests summarized thinking on streaming runs. | `claude-fable-5-1`, `claude-fable-5`, `claude-sonnet-5`, `claude-opus-5`, `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-opus-4-5`, `claude-sonnet-4-5`, `claude-haiku-4-5` |
-| Azure OpenAI | Summaries only — Requires OpenAI organization verification; falls back to no summaries. | `azure/gpt-5.4`, `azure/gpt-5.4-mini`, `azure/gpt-5.4-nano`, `azure/gpt-5.2`, `azure/gpt-5.1`, `azure/gpt-5.1-codex`, `azure/gpt-5`, `azure/gpt-5-mini`, `azure/gpt-5-nano`, `azure/o3`, `azure/o4-mini` |
-| Azure Anthropic | Summaries only — These generations omit full thinking; Sim requests summarized thinking on streaming runs. | `azure-anthropic/claude-opus-4-6`, `azure-anthropic/claude-opus-4-5`, `azure-anthropic/claude-sonnet-4-5`, `azure-anthropic/claude-opus-4-1`, `azure-anthropic/claude-haiku-4-5` |
-| Google | Summaries only | `gemini-3.8-flash`, `gemini-3.6-flash`, `gemini-3.5-flash-lite`, `gemini-3.5-flash`, `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite`, `gemini-3-flash-preview`, `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite` |
-| Vertex AI | Summaries only | `vertex/gemini-3.5-flash`, `vertex/gemini-3.1-pro-preview`, `vertex/gemini-3.1-flash-lite`, `vertex/gemini-3-flash-preview`, `vertex/gemini-2.5-pro`, `vertex/gemini-2.5-flash`, `vertex/gemini-2.5-flash-lite` |
-| DeepSeek | Full thinking deltas | `deepseek-v4-pro`, `deepseek-v4-flash`, `deepseek-reasoner` |
+| Azure OpenAI | Summaries only — Requires OpenAI organization verification; falls back to no summaries. | `azure/gpt-6-astra`, `azure/gpt-5.6-sol`, `azure/gpt-5.6-terra`, `azure/gpt-5.6-luna`, `azure/gpt-5.5`, `azure/gpt-5.4-pro`, `azure/gpt-5.4`, `azure/gpt-5.4-mini`, `azure/gpt-5.4-nano`, `azure/gpt-5.2`, `azure/gpt-5.1`, `azure/gpt-5.1-codex`, `azure/gpt-5`, `azure/gpt-5-mini`, `azure/gpt-5-nano`, `azure/o3`, `azure/o4-mini` |
+| Azure Anthropic | Summaries only — These generations omit full thinking; Sim requests summarized thinking on streaming runs. | `azure-anthropic/claude-fable-5-1`, `azure-anthropic/claude-opus-5`, `azure-anthropic/claude-opus-4-8`, `azure-anthropic/claude-opus-4-7`, `azure-anthropic/claude-opus-4-6`, `azure-anthropic/claude-opus-4-5`, `azure-anthropic/claude-sonnet-5`, `azure-anthropic/claude-sonnet-4-6`, `azure-anthropic/claude-sonnet-4-5`, `azure-anthropic/claude-opus-4-1`, `azure-anthropic/claude-haiku-4-5` |
+| Google | Summaries only | `gemini-3.8-flash`, `gemini-3.7-flash`, `gemini-3.6-flash`, `gemini-3.5-flash-lite`, `gemini-3.5-flash`, `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite`, `gemini-3-flash-preview`, `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite` |
+| Vertex AI | Summaries only | `vertex/gemini-3.8-flash`, `vertex/gemini-3.7-flash`, `vertex/gemini-3.6-flash`, `vertex/gemini-3.5-flash-lite`, `vertex/gemini-3.5-flash`, `vertex/gemini-3.1-pro-preview`, `vertex/gemini-3.1-flash-lite`, `vertex/gemini-3-flash-preview`, `vertex/gemini-2.5-pro`, `vertex/gemini-2.5-flash`, `vertex/gemini-2.5-flash-lite` |
+| DeepSeek | Full thinking deltas | `deepseek-v4-pro`, `deepseek-flash`, `deepseek-v4-flash`, `deepseek-reasoner` |
 | xAI | Full thinking deltas | `grok-4.6`, `grok-4.5`, `grok-4.3`, `grok-4.20-multi-agent-0309` |
-| Groq | Full thinking deltas | `groq/openai/gpt-oss-120b`, `groq/openai/gpt-oss-20b`, `groq/openai/gpt-oss-safeguard-20b`, `groq/qwen/qwen3.6-27b` |
+| Cerebras | Full thinking deltas | `cerebras/qwen-3.8-27b` |
+| Groq | Full thinking deltas | `groq/openai/gpt-oss-120b`, `groq/openai/gpt-oss-20b`, `groq/openai/gpt-oss-safeguard-20b`, `groq/qwen/qwen3.8-27b`, `groq/qwen/qwen3.6-27b` |
 | Meta | Not streamed | `muse-spark-1.3`, `muse-spark-1.1` |
-| Kimi | Full thinking deltas | `kimi-k2.6` |
-| Z.ai | Full thinking deltas | `glm-5.3`, `glm-5.2`, `glm-5.1`, `glm-5`, `glm-5-turbo`, `glm-4.7`, `glm-4.6`, `glm-4.5`, `glm-4.5-air` |
+| Kimi | Full thinking deltas | `kimi-k3`, `kimi-k2.6` |
+| Z.ai | Full thinking deltas | `glm-5.3`, `glm-5.3-flash`, `glm-5.2`, `glm-5.1`, `glm-5`, `glm-5-turbo`, `glm-4.7`, `glm-4.6`, `glm-4.5`, `glm-4.5-air` |
 
 {/* agent-stream-capabilities:end */}
 
diff --git a/apps/sim/app/api/providers/ollama-cloud/models/route.test.ts b/apps/sim/app/api/providers/ollama-cloud/models/route.test.ts
index a4b4cbfedda..3c6d2537159 100644
--- a/apps/sim/app/api/providers/ollama-cloud/models/route.test.ts
+++ b/apps/sim/app/api/providers/ollama-cloud/models/route.test.ts
@@ -144,6 +144,29 @@ describe('GET /api/providers/ollama-cloud/models', () => {
     expect(fetchAuthHeader()).toBe('Bearer byok-ollama-key')
   })
 
+  it('discovers newly available models on subsequent requests without a static catalog update', async () => {
+    grantWorkspaceAccess()
+    mockGetBYOKKey.mockResolvedValue({ apiKey: 'byok-ollama-key' })
+    mockFetch
+      .mockResolvedValueOnce(okResponse({ models: [{ name: 'kimi-k3' }] }))
+      .mockResolvedValueOnce(
+        okResponse({
+          models: [{ name: 'kimi-k3' }, { name: 'deepseek-v4.1-flash' }, { name: 'glm-5.3' }],
+        })
+      )
+
+    const first = await GET(requestWithWorkspace('ws-1'))
+    expect(await first.json()).toEqual({ models: ['ollama-cloud/kimi-k3'] })
+    const refreshed = await GET(requestWithWorkspace('ws-1'))
+    expect(await refreshed.json()).toEqual({
+      models: ['ollama-cloud/kimi-k3', 'ollama-cloud/deepseek-v4.1-flash', 'ollama-cloud/glm-5.3'],
+    })
+    expect(mockFetch).toHaveBeenLastCalledWith(
+      OLLAMA_CLOUD_TAGS_URL,
+      expect.objectContaining({ cache: 'no-store' })
+    )
+  })
+
   it('does not call getBYOKKey when there is a workspaceId but no session', async () => {
     mockGetSession.mockResolvedValue(null)
 
diff --git a/apps/sim/blocks/provider-credentials.test.ts b/apps/sim/blocks/provider-credentials.test.ts
new file mode 100644
index 00000000000..dcaaee7aceb
--- /dev/null
+++ b/apps/sim/blocks/provider-credentials.test.ts
@@ -0,0 +1,34 @@
+/**
+ * @vitest-environment node
+ */
+import { describe, expect, it } from 'vitest'
+import { evaluateSubBlockCondition } from '@/lib/workflows/subblocks/visibility'
+import { getProviderCredentialSubBlocks } from '@/blocks/utils'
+
+describe('provider credential visibility', () => {
+  const subBlocks = getProviderCredentialSubBlocks().filter(({ id }) => id !== 'apiKey')
+
+  it.each([
+    ['azure/MyDeployment', ['azureEndpoint', 'azureApiVersion']],
+    ['AZURE/MyDeployment', ['azureEndpoint', 'azureApiVersion']],
+    ['azure-anthropic/MyDeployment', ['azureEndpoint', 'azureApiVersion']],
+    ['bedrock/custom-profile', ['bedrockAccessKeyId', 'bedrockSecretKey', 'bedrockRegion']],
+    [
+      'vertex/publishers/google/models/custom-gemini',
+      ['vertexCredential', 'vertexManualCredential', 'vertexProject', 'vertexLocation'],
+    ],
+    [
+      'VERTEX/CustomModel',
+      ['vertexCredential', 'vertexManualCredential', 'vertexProject', 'vertexLocation'],
+    ],
+    ['gpt-4o', []],
+    ['unknown/model', []],
+    ['', []],
+  ])('shows only the routed provider credentials for %s', (model, expected) => {
+    const visible = subBlocks
+      .filter((subBlock) => evaluateSubBlockCondition(subBlock.condition, { model }))
+      .map(({ id }) => id)
+
+    expect(visible).toEqual(expected)
+  })
+})
diff --git a/apps/sim/blocks/utils.test.ts b/apps/sim/blocks/utils.test.ts
index 1cd89fa4229..c260de06e87 100644
--- a/apps/sim/blocks/utils.test.ts
+++ b/apps/sim/blocks/utils.test.ts
@@ -31,7 +31,8 @@ const { mockProviders } = vi.hoisted(() => ({
   },
 }))
 
-vi.mock('@/providers/models', () => ({
+vi.mock('@/providers/models', async (importOriginal) => ({
+  ...(await importOriginal<typeof import('@/providers/models')>()),
   getProviderFileAttachment: vi
     .fn()
     .mockReturnValue({ maxBytes: 10 * 1024 * 1024, strategy: 'inline' }),
@@ -180,6 +181,16 @@ describe('getApiKeyCondition / shouldRequireApiKeyForModel', () => {
   })
 
   describe('provider store lookup (client-side)', () => {
+    it('requires the cloud key even when a local discovered name uses its namespace', () => {
+      mockProviders.value.ollama.models = ['azure/MyDeployment', 'ollama-cloud/MyModel']
+      expect(evaluateCondition('azure/MyDeployment')).toBe(true)
+      expect(evaluateCondition('ollama-cloud/MyModel')).toBe(true)
+    })
+
+    it('does not require an API key for an undiscovered namespaced Ollama model', () => {
+      expect(evaluateCondition('OLLAMA/Org/CustomModel')).toBe(false)
+    })
+
     it('does not require API key when model is in the Ollama store bucket', () => {
       mockProviders.value.ollama.models = ['llama3:latest', 'mistral:latest']
       expect(evaluateCondition('llama3:latest')).toBe(false)
diff --git a/apps/sim/blocks/utils.ts b/apps/sim/blocks/utils.ts
index dca2389b503..bc81c39cc2a 100644
--- a/apps/sim/blocks/utils.ts
+++ b/apps/sim/blocks/utils.ts
@@ -6,12 +6,14 @@ import { getScopesForService } from '@/lib/oauth/utils'
 import { containsReference } from '@/lib/workflows/sanitization/references'
 import type { SubBlockConfig } from '@/blocks/types'
 import {
+  findProviderFromModel,
   getBaseModelProviders,
   getHostedModels,
   getModelSunsetStatus,
   getProviderIcon,
   getProviderModels,
   isAutoModel,
+  isCustomModelId,
   orderModelIdsByReleaseDate,
   SIM_AUTO_MODEL_ID,
 } from '@/providers/models'
@@ -20,8 +22,6 @@ import type { ProviderId } from '@/providers/types'
 import { getProviderFromModel } from '@/providers/utils'
 import { useProvidersStore } from '@/stores/providers/store'
 
-export const VERTEX_MODELS = getProviderModels('vertex')
-export const BEDROCK_MODELS = getProviderModels('bedrock')
 export const AZURE_MODELS = [
   ...getProviderModels('azure-openai'),
   ...getProviderModels('azure-anthropic'),
@@ -164,10 +164,16 @@ function shouldRequireApiKeyForModel(model: string): boolean {
   ) {
     return false
   }
-  if (normalizedModel.startsWith('vllm/') || normalizedModel.startsWith('litellm/')) {
+  if (
+    normalizedModel.startsWith('ollama/') ||
+    normalizedModel.startsWith('vllm/') ||
+    normalizedModel.startsWith('litellm/')
+  ) {
     return false
   }
 
+  if (isCustomModelId(normalizedModel)) return true
+
   const storeProvider = getProviderFromStore(normalizedModel)
   if (storeProvider === 'ollama' || storeProvider === 'vllm' || storeProvider === 'litellm')
     return false
@@ -272,6 +278,14 @@ export function getCohereRerankerApiKeyCondition() {
   }
 }
 
+function getModelProviderCondition(...providerIds: ProviderId[]) {
+  return (values?: Record<string, unknown>) => {
+    const model = typeof values?.model === 'string' ? values.model : ''
+    const provider = findProviderFromModel(model.trim())
+    return buildModelVisibilityCondition(model, provider !== null && providerIds.includes(provider))
+  }
+}
+
 /**
  * Returns the standard provider credential subblocks used by LLM-based blocks.
  * This includes: Vertex AI OAuth, API Key, Azure (OpenAI + Anthropic), Vertex AI config, and Bedrock config.
@@ -290,10 +304,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       requiredScopes: getScopesForService('vertex-ai'),
       placeholder: 'Select Google Cloud account',
       required: true,
-      condition: {
-        field: 'model',
-        value: VERTEX_MODELS,
-      },
+      condition: getModelProviderCondition('vertex'),
     },
     {
       id: 'vertexManualCredential',
@@ -303,10 +314,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       mode: 'advanced',
       placeholder: 'Enter credential ID',
       required: true,
-      condition: {
-        field: 'model',
-        value: VERTEX_MODELS,
-      },
+      condition: getModelProviderCondition('vertex'),
     },
     {
       id: 'apiKey',
@@ -326,10 +334,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       placeholder: 'https://your-resource.services.ai.azure.com',
       connectionDroppable: false,
       hideWhenEnvSet: 'NEXT_PUBLIC_AZURE_CONFIGURED',
-      condition: {
-        field: 'model',
-        value: AZURE_MODELS,
-      },
+      condition: getModelProviderCondition('azure-openai', 'azure-anthropic'),
     },
     {
       id: 'azureApiVersion',
@@ -338,10 +343,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       placeholder: 'Enter API version',
       connectionDroppable: false,
       hideWhenEnvSet: 'NEXT_PUBLIC_AZURE_CONFIGURED',
-      condition: {
-        field: 'model',
-        value: AZURE_MODELS,
-      },
+      condition: getModelProviderCondition('azure-openai', 'azure-anthropic'),
     },
     {
       id: 'vertexProject',
@@ -351,10 +353,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       placeholder: 'your-gcp-project-id',
       connectionDroppable: false,
       required: true,
-      condition: {
-        field: 'model',
-        value: VERTEX_MODELS,
-      },
+      condition: getModelProviderCondition('vertex'),
     },
     {
       id: 'vertexLocation',
@@ -363,10 +362,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       placeholder: 'us-central1',
       connectionDroppable: false,
       required: true,
-      condition: {
-        field: 'model',
-        value: VERTEX_MODELS,
-      },
+      condition: getModelProviderCondition('vertex'),
     },
     {
       id: 'bedrockAccessKeyId',
@@ -377,10 +373,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       connectionDroppable: false,
       required: true,
       hideWhenEnvSet: 'NEXT_PUBLIC_BEDROCK_DEFAULT_CREDENTIALS',
-      condition: {
-        field: 'model',
-        value: BEDROCK_MODELS,
-      },
+      condition: getModelProviderCondition('bedrock'),
     },
     {
       id: 'bedrockSecretKey',
@@ -391,10 +384,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       connectionDroppable: false,
       required: true,
       hideWhenEnvSet: 'NEXT_PUBLIC_BEDROCK_DEFAULT_CREDENTIALS',
-      condition: {
-        field: 'model',
-        value: BEDROCK_MODELS,
-      },
+      condition: getModelProviderCondition('bedrock'),
     },
     {
       id: 'bedrockRegion',
@@ -402,10 +392,7 @@ export function getProviderCredentialSubBlocks(): SubBlockConfig[] {
       type: 'short-input',
       placeholder: 'us-east-1',
       connectionDroppable: false,
-      condition: {
-        field: 'model',
-        value: BEDROCK_MODELS,
-      },
+      condition: getModelProviderCondition('bedrock'),
     },
   ]
 }
diff --git a/apps/sim/lib/api-key/byok.test.ts b/apps/sim/lib/api-key/byok.test.ts
index abba750dbac..5a905237cf2 100644
--- a/apps/sim/lib/api-key/byok.test.ts
+++ b/apps/sim/lib/api-key/byok.test.ts
@@ -451,6 +451,99 @@ describe('getBYOKKey', () => {
   })
 })
 
+describe('getApiKeyWithBYOK provider classification', () => {
+  const dynamicProviders = [
+    'ollama',
+    'vllm',
+    'litellm',
+    'fireworks',
+    'together',
+    'baseten',
+    'ollama-cloud',
+  ] as const
+
+  beforeEach(() => {
+    vi.clearAllMocks()
+    resetDbChainMock()
+    mockIsHosted.value = true
+    mockEnv.AZURE_OPENAI_API_KEY = 'azure-env-key'
+    mockEnv.AZURE_ANTHROPIC_API_KEY = 'azure-anthropic-env-key'
+    mockEnv.VLLM_API_KEY = 'vllm-env-key'
+    mockEnv.LITELLM_API_KEY = 'litellm-env-key'
+    dbChainMockFns.orderBy.mockResolvedValue([storedKey('other-provider-key')])
+    mockDecryptSecret.mockImplementation(async (encrypted: string) => ({
+      decrypted: encrypted.replace('encrypted-', 'decrypted-'),
+    }))
+  })
+
+  it.each(dynamicProviders)(
+    'keeps Azure credentials when %s discovery contains the same model ID',
+    async (discoveredProvider) => {
+      const model = 'AZURE/CustomDeployment'
+      vi.mocked(useProvidersStore.getState).mockReturnValue({
+        providers: Object.fromEntries(
+          dynamicProviders.map((provider) => [
+            provider,
+            { models: provider === discoveredProvider ? [model] : [] },
+          ])
+        ),
+      } as ReturnType<typeof useProvidersStore.getState>)
+
+      const result = await getApiKeyWithBYOK('azure-openai', model, uniqueWorkspaceId())
+
+      expect(result).toEqual({ apiKey: 'azure-env-key', isBYOK: false })
+      expect(dbChainMockFns.where).not.toHaveBeenCalled()
+      expect(mockGetRotatingApiKey).not.toHaveBeenCalled()
+    }
+  )
+
+  it.each([
+    ['vertex', 'vertex/CustomDeployment', 'vertex-access-token'],
+    ['azure-anthropic', 'azure-anthropic/CustomDeployment', 'azure-anthropic-user-key'],
+  ])(
+    'retains caller credentials for %s despite a local model name collision',
+    async (provider, model, apiKey) => {
+      vi.mocked(useProvidersStore.getState).mockReturnValue({
+        providers: Object.fromEntries(dynamicProviders.map((name) => [name, { models: [model] }])),
+      } as ReturnType<typeof useProvidersStore.getState>)
+
+      expect(await getApiKeyWithBYOK(provider, model, uniqueWorkspaceId(), apiKey)).toEqual({
+        apiKey,
+        isBYOK: false,
+      })
+      expect(dbChainMockFns.where).not.toHaveBeenCalled()
+    }
+  )
+
+  it.each([
+    ['ollama', 'empty'],
+    ['vllm', 'vllm-env-key'],
+    ['litellm', 'litellm-env-key'],
+  ])('preserves %s authentication for a custom unprefixed model', async (provider, apiKey) => {
+    expect(await getApiKeyWithBYOK(provider, 'MyCustomModel', uniqueWorkspaceId())).toEqual({
+      apiKey,
+      isBYOK: false,
+    })
+    expect(dbChainMockFns.where).not.toHaveBeenCalled()
+  })
+
+  it.each(['vllm', 'litellm'])(
+    'prefers a caller key to the configured %s key for a local model',
+    async (provider) => {
+      expect(
+        await getApiKeyWithBYOK(provider, 'MyCustomModel', uniqueWorkspaceId(), 'caller-key')
+      ).toEqual({ apiKey: 'caller-key', isBYOK: false })
+    }
+  )
+
+  it('uses Bedrock credentials for an uncataloged inference profile', async () => {
+    expect(
+      await getApiKeyWithBYOK('bedrock', 'BEDROCK/MyInferenceProfile', uniqueWorkspaceId())
+    ).toEqual({ apiKey: 'placeholder', isBYOK: false })
+    expect(dbChainMockFns.where).not.toHaveBeenCalled()
+  })
+})
+
 describe('getApiKeyWithBYOK for Fireworks', () => {
   const HOSTED_POOL_MODEL = 'fireworks/glm-5.2'
 
diff --git a/apps/sim/lib/api-key/byok.ts b/apps/sim/lib/api-key/byok.ts
index 1bacd8af607..a9ce7d43e08 100644
--- a/apps/sim/lib/api-key/byok.ts
+++ b/apps/sim/lib/api-key/byok.ts
@@ -10,7 +10,6 @@ import { isHosted } from '@/lib/core/config/env-flags'
 import { decryptSecret } from '@/lib/core/security/encryption'
 import { getHostedModels } from '@/providers/models'
 import { PROVIDER_PLACEHOLDER_KEY } from '@/providers/utils'
-import { useProvidersStore } from '@/stores/providers/store'
 import type { BYOKProviderId } from '@/tools/types'
 
 const logger = createLogger('BYOKKeys')
@@ -183,6 +182,9 @@ export async function getBYOKKey(
 }
 
 /**
+ * Resolves credentials for the provider already selected by model routing.
+ * Discovery lists must not override that provider and select a different key pool.
+ *
  * `scope` is present only when the key came from a stored BYOK pool; a
  * Sim-hosted, env, or caller-supplied key has no scope. Declared rather than
  * dropped so the returned type matches what a BYOK branch actually hands back.
@@ -193,28 +195,19 @@ export async function getApiKeyWithBYOK(
   workspaceId: string | undefined | null,
   userProvidedKey?: string
 ): Promise<{ apiKey: string; isBYOK: boolean; scope?: BYOKKeyScopeName }> {
-  const isOllamaModel =
-    provider === 'ollama' || useProvidersStore.getState().providers.ollama.models.includes(model)
-  if (isOllamaModel) {
+  if (provider === 'ollama') {
     return { apiKey: 'empty', isBYOK: false }
   }
 
-  const isVllmModel =
-    provider === 'vllm' || useProvidersStore.getState().providers.vllm.models.includes(model)
-  if (isVllmModel) {
+  if (provider === 'vllm') {
     return { apiKey: userProvidedKey || env.VLLM_API_KEY || 'empty', isBYOK: false }
   }
 
-  const isLitellmModel =
-    provider === 'litellm' || useProvidersStore.getState().providers.litellm.models.includes(model)
-  if (isLitellmModel) {
+  if (provider === 'litellm') {
     return { apiKey: userProvidedKey || env.LITELLM_API_KEY || 'empty', isBYOK: false }
   }
 
-  const isFireworksModel =
-    provider === 'fireworks' ||
-    useProvidersStore.getState().providers.fireworks.models.includes(model)
-  if (isFireworksModel) {
+  if (provider === 'fireworks') {
     if (workspaceId) {
       const byokResult = await getBYOKKey(workspaceId, 'fireworks')
       if (byokResult) {
@@ -260,10 +253,7 @@ export async function getApiKeyWithBYOK(
     throw new Error(`API key is required for Fireworks ${model}`)
   }
 
-  const isTogetherModel =
-    provider === 'together' ||
-    useProvidersStore.getState().providers.together.models.includes(model)
-  if (isTogetherModel) {
+  if (provider === 'together') {
     if (workspaceId) {
       const byokResult = await getBYOKKey(workspaceId, 'together')
       if (byokResult) {
@@ -284,9 +274,7 @@ export async function getApiKeyWithBYOK(
     throw new Error(`API key is required for Together AI ${model}`)
   }
 
-  const isBasetenModel =
-    provider === 'baseten' || useProvidersStore.getState().providers.baseten.models.includes(model)
-  if (isBasetenModel) {
+  if (provider === 'baseten') {
     if (workspaceId) {
       const byokResult = await getBYOKKey(workspaceId, 'baseten')
       if (byokResult) {
@@ -303,10 +291,7 @@ export async function getApiKeyWithBYOK(
     throw new Error(`API key is required for Baseten ${model}`)
   }
 
-  const isOllamaCloudModel =
-    provider === 'ollama-cloud' ||
-    useProvidersStore.getState().providers['ollama-cloud'].models.includes(model)
-  if (isOllamaCloudModel) {
+  if (provider === 'ollama-cloud') {
     if (workspaceId) {
       const byokResult = await getBYOKKey(workspaceId, 'ollama-cloud')
       if (byokResult) {
@@ -324,8 +309,7 @@ export async function getApiKeyWithBYOK(
     throw new Error(`API key is required for Ollama Cloud ${model}`)
   }
 
-  const isBedrockModel = provider === 'bedrock' || model.startsWith('bedrock/')
-  if (isBedrockModel) {
+  if (provider === 'bedrock') {
     return { apiKey: PROVIDER_PLACEHOLDER_KEY, isBYOK: false }
   }
 
diff --git a/apps/sim/lib/core/security/input-validation.test.ts b/apps/sim/lib/core/security/input-validation.test.ts
index d86b7d87d49..7c06184c525 100644
--- a/apps/sim/lib/core/security/input-validation.test.ts
+++ b/apps/sim/lib/core/security/input-validation.test.ts
@@ -1402,6 +1402,8 @@ describe('validateGoogleCloudLocation', () => {
       'africa-south1',
       'me-central2',
       'global',
+      'us',
+      'eu',
     ])('should accept %s', (location) => {
       const result = validateGoogleCloudLocation(location)
       expect(result.isValid).toBe(true)
diff --git a/apps/sim/lib/core/security/input-validation.ts b/apps/sim/lib/core/security/input-validation.ts
index 9aeb1b6eb19..4a1e460d19f 100644
--- a/apps/sim/lib/core/security/input-validation.ts
+++ b/apps/sim/lib/core/security/input-validation.ts
@@ -661,7 +661,7 @@ export function validateAwsRegion(
  * and relocate the request — along with any attached credential — to an
  * attacker-controlled host.
  *
- * Accepts `global` plus the documented `{geography}-{direction}{index}` region
+ * Accepts `global`, the `us` and `eu` multi-regions, and the `{geography}-{direction}{index}` region
  * form (e.g. us-central1, europe-west4, northamerica-northeast1, me-central2).
  *
  * @param value - The location to validate
@@ -677,7 +677,7 @@ export function validateGoogleCloudLocation(
   }
 
   const googleLocationPattern =
-    /^(global|(africa|asia|australia|europe|me|northamerica|southamerica|us)-(central|east|north|northeast|northwest|south|southeast|southwest|west)\d{1,2})$/
+    /^(global|us|eu|(africa|asia|australia|europe|me|northamerica|southamerica|us)-(central|east|north|northeast|northwest|south|southeast|southwest|west)\d{1,2})$/
 
   if (!googleLocationPattern.test(value)) {
     logger.warn('Invalid Google Cloud location format', {
diff --git a/apps/sim/lib/workflows/editing/validation.test.ts b/apps/sim/lib/workflows/editing/validation.test.ts
index bd1a65a2d89..1f192cb1721 100644
--- a/apps/sim/lib/workflows/editing/validation.test.ts
+++ b/apps/sim/lib/workflows/editing/validation.test.ts
@@ -493,6 +493,40 @@ describe('validateInputsForBlock', () => {
     expect(result.validInputs.model).toBe('ollama/my-private-model')
   })
 
+  it.each([
+    'azure/MyDeployment',
+    'AZURE/MyDeployment',
+    'azure-anthropic/MyDeployment',
+    'bedrock/custom-inference-profile',
+    'vertex/publishers/google/models/custom-gemini',
+    'GROQ/Org/CustomModel',
+    'CEREBRAS/CustomModel',
+    'NVIDIA/CustomModel',
+  ])('accepts a custom cloud model ID: %s', (model) => {
+    for (const blockType of ['agent', 'router_v2']) {
+      const result = validateInputsForBlock(blockType, { model: `  ${model}  ` }, 'block-1')
+      expect(result.errors).toEqual([])
+      expect(result.validInputs.model).toBe(model)
+    }
+  })
+
+  it.each([
+    'azure/',
+    'azure-anthropic/',
+    'bedrock/',
+    'vertex/',
+    'groq/',
+    'cerebras/',
+    'nvidia/',
+    'ollama/',
+    'ollama-cloud/',
+    'unknown/model',
+  ])('rejects incomplete or unsupported cloud namespaces: %s', (model) => {
+    const result = validateInputsForBlock('agent', { model }, 'agent-1')
+    expect(result.validInputs.model).toBeUndefined()
+    expect(result.errors[0]?.error).toContain('Unknown model id')
+  })
+
   it('validates the model field on router_v2 blocks too', () => {
     const valid = validateInputsForBlock('router_v2', { model: 'claude-sonnet-4-6' }, 'router-1')
     expect(valid.errors).toHaveLength(0)
diff --git a/apps/sim/lib/workflows/editing/validation.ts b/apps/sim/lib/workflows/editing/validation.ts
index ec2a366a1bd..e02ca16ef02 100644
--- a/apps/sim/lib/workflows/editing/validation.ts
+++ b/apps/sim/lib/workflows/editing/validation.ts
@@ -24,7 +24,12 @@ import type { SubBlockConfig } from '@/blocks/types'
 import { getModelOptions } from '@/blocks/utils'
 import { overlayVisibility } from '@/blocks/visibility/context'
 import { BlockType, EDGE, normalizeName } from '@/executor/constants'
-import { isAutoModel, isKnownModelId, suggestModelIdsForUnknownModel } from '@/providers/models'
+import {
+  isAutoModel,
+  isCustomModelId,
+  isKnownModelId,
+  suggestModelIdsForUnknownModel,
+} from '@/providers/models'
 import { isPiByokOnlyMode } from '@/providers/pi-providers'
 import { getTool } from '@/tools/utils'
 import {
@@ -680,7 +685,7 @@ export function validateValueForSubBlockType(
         if (trimmed !== '' && isAutoModel(trimmed) && isHostedDeployment) {
           return { valid: true, value: trimmed.toLowerCase() }
         }
-        if (trimmed !== '' && !isKnownModelId(trimmed)) {
+        if (trimmed !== '' && !isKnownModelId(trimmed) && !isCustomModelId(trimmed)) {
           const suggestions = suggestModelIdsForUnknownModel(trimmed)
           const suggestionText =
             suggestions.length > 0 ? ` Valid options include: ${suggestions.join(', ')}.` : ''
@@ -691,7 +696,7 @@ export function validateValueForSubBlockType(
               blockType,
               field: fieldName,
               value,
-              error: `Unknown model id "${trimmed}" for block "${blockType}". Read components/blocks/${blockType}.json (the model.options array) for valid ids; prefer entries with recommended: true and avoid deprecated: true. For user-configured models (Ollama, Ollama Cloud, vLLM, LiteLLM, OpenRouter, Fireworks, Together AI, Baseten), prefix the id with the provider slash, e.g. "ollama/llama3.1:8b" or "ollama-cloud/gpt-oss:120b".${suggestionText}`,
+              error: `Unknown model id "${trimmed}" for block "${blockType}". Read components/blocks/${blockType}.json (the model.options array) for valid ids; prefer entries with recommended: true and avoid deprecated: true. For user-configured models, use a supported provider namespace, e.g. "azure/my-deployment", "azure-anthropic/my-deployment", "bedrock/my-inference-profile", "vertex/my-model", "ollama/llama3.1:8b", or "openrouter/provider/model".${suggestionText}`,
             },
           }
         }
diff --git a/apps/sim/providers/azure-anthropic/index.test.ts b/apps/sim/providers/azure-anthropic/index.test.ts
index 1de56f6ed15..cb0c777c82b 100644
--- a/apps/sim/providers/azure-anthropic/index.test.ts
+++ b/apps/sim/providers/azure-anthropic/index.test.ts
@@ -110,6 +110,17 @@ describe('azureAnthropicProvider — SSRF pinning', () => {
     expect(buildClientOptions().defaultHeaders).not.toHaveProperty('anthropic-beta')
   })
 
+  it('preserves custom deployment casing when removing an uppercase routing prefix', async () => {
+    setEnv({ AZURE_ANTHROPIC_ENDPOINT: 'https://custom.services.ai.azure.com' })
+    const providerRequest = request({ model: 'AZURE-ANTHROPIC/Team-Claude-Deployment' })
+
+    await azureAnthropicProvider.executeRequest(providerRequest)
+
+    const [forwardedRequest, config] = mockExecuteAnthropic.mock.calls[0]
+    expect(forwardedRequest.model).toBe('AZURE-ANTHROPIC/Team-Claude-Deployment')
+    expect(config.resolveWireModel(forwardedRequest)).toBe('Team-Claude-Deployment')
+  })
+
   it('throws and never builds a client when validation blocks the endpoint', async () => {
     mockValidate.mockResolvedValue({ isValid: false, error: 'resolves to a blocked IP address' })
 
diff --git a/apps/sim/providers/azure-anthropic/index.ts b/apps/sim/providers/azure-anthropic/index.ts
index dff9f1be274..8431204cd73 100644
--- a/apps/sim/providers/azure-anthropic/index.ts
+++ b/apps/sim/providers/azure-anthropic/index.ts
@@ -64,7 +64,7 @@ export const azureAnthropicProvider: ProviderConfig = {
     return executeAnthropicProviderRequest(request, {
       providerId: 'azure-anthropic',
       providerLabel: 'Azure Anthropic',
-      resolveWireModel: ({ model }) => model.replace(/^azure-anthropic\//, ''),
+      resolveWireModel: ({ model }) => model.replace(/^azure-anthropic\//i, ''),
       createClient: (apiKey) => {
         const cacheKey = [
           'azure-anthropic',
diff --git a/apps/sim/providers/azure-openai/index.test.ts b/apps/sim/providers/azure-openai/index.test.ts
index 48c7431cb36..bf64de8121d 100644
--- a/apps/sim/providers/azure-openai/index.test.ts
+++ b/apps/sim/providers/azure-openai/index.test.ts
@@ -163,6 +163,20 @@ describe('azureOpenAIProvider — SSRF pinning', () => {
       expect(responsesConfig().fetch).toBeUndefined()
     })
 
+    it.each([false, true])(
+      'preserves a custom deployment name through Responses routing (full endpoint: %s)',
+      async (fullEndpoint) => {
+        mockIsResponsesEndpoint.mockReturnValue(fullEndpoint)
+        setEnv({ AZURE_OPENAI_ENDPOINT: 'https://custom.openai.azure.com' })
+        const providerRequest = request({ model: 'AZURE/Team-GPT-Deployment' })
+
+        await azureOpenAIProvider.executeRequest(providerRequest)
+
+        expect(mockExecuteResponses.mock.calls[0][0].model).toBe('AZURE/Team-GPT-Deployment')
+        expect(responsesConfig().modelName).toBe('Team-GPT-Deployment')
+      }
+    )
+
     it('throws and never reaches the Responses core when validation blocks the endpoint', async () => {
       mockValidate.mockResolvedValue({ isValid: false, error: 'resolves to a blocked IP address' })
 
@@ -215,6 +229,21 @@ describe('azureOpenAIProvider — SSRF pinning', () => {
       expect(azureOpenAIArgs[0]).not.toHaveProperty('fetch')
     })
 
+    it('preserves a custom deployment name through Chat Completions routing', async () => {
+      mockIsChatCompletionsEndpoint.mockReturnValue(true)
+      setEnv({
+        AZURE_OPENAI_ENDPOINT: 'https://custom.openai.azure.com/openai/v1/chat/completions',
+      })
+      mockChatCreate.mockResolvedValue({
+        choices: [{ message: { content: 'hi' } }],
+        usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
+      })
+
+      await azureOpenAIProvider.executeRequest(request({ model: 'AZURE/Team-GPT-Deployment' }))
+
+      expect(mockChatCreate.mock.calls[0][0].model).toBe('Team-GPT-Deployment')
+    })
+
     it('projects the settled tool-loop answer without a final streaming request', async () => {
       mockIsChatCompletionsEndpoint.mockReturnValue(true)
       mockValidate.mockResolvedValue({ isValid: true, resolvedIP: '203.0.113.10' })
diff --git a/apps/sim/providers/azure-openai/index.ts b/apps/sim/providers/azure-openai/index.ts
index 72251763d99..5905b59f4f2 100644
--- a/apps/sim/providers/azure-openai/index.ts
+++ b/apps/sim/providers/azure-openai/index.ts
@@ -703,7 +703,7 @@ export const azureOpenAIProvider: ProviderConfig = {
 
       // Try to extract deployment from URL, fall back to model name
       const urlDeployment = extractDeploymentFromUrl(azureEndpoint)
-      const deploymentName = urlDeployment || request.model.replace('azure/', '')
+      const deploymentName = urlDeployment || request.model.replace(/^azure\//i, '')
 
       // Try to extract api-version from URL, fall back to request param or env or default
       const urlApiVersion = extractApiVersionFromUrl(azureEndpoint)
@@ -733,7 +733,7 @@ export const azureOpenAIProvider: ProviderConfig = {
     if (isResponsesEndpoint(azureEndpoint)) {
       logger.info('Detected full responses endpoint URL, using it directly')
 
-      const deploymentName = request.model.replace('azure/', '')
+      const deploymentName = request.model.replace(/^azure\//i, '')
 
       // Use the URL as-is since it's already complete
       return executeResponsesProviderRequest(
@@ -758,7 +758,7 @@ export const azureOpenAIProvider: ProviderConfig = {
     logger.info('Using base endpoint, constructing Responses API URL')
     const azureApiVersion =
       request.azureApiVersion || env.AZURE_OPENAI_API_VERSION || '2024-07-01-preview'
-    const deploymentName = request.model.replace('azure/', '')
+    const deploymentName = request.model.replace(/^azure\//i, '')
     const apiUrl = `${azureEndpoint.replace(/\/$/, '')}/openai/v1/responses?api-version=${azureApiVersion}`
 
     return executeResponsesProviderRequest(
diff --git a/apps/sim/providers/baseten/index.test.ts b/apps/sim/providers/baseten/index.test.ts
index d0a4ed0308c..2dabec8909e 100644
--- a/apps/sim/providers/baseten/index.test.ts
+++ b/apps/sim/providers/baseten/index.test.ts
@@ -155,6 +155,14 @@ describe('basetenProvider', () => {
     await expect(basetenProvider.executeRequest(baseRequest)).rejects.toBeInstanceOf(ProviderError)
   })
 
+  it('preserves custom model casing after an uppercase provider prefix', async () => {
+    mockCreate.mockResolvedValueOnce(textResponse('ok'))
+
+    await basetenProvider.executeRequest({ ...baseRequest, model: 'BASETEN/Org/Custom-Model' })
+
+    expect(callBody(0).model).toBe('Org/Custom-Model')
+  })
+
   it('streams directly when there are no tools', async () => {
     mockCreate.mockResolvedValueOnce({})
 
diff --git a/apps/sim/providers/baseten/index.ts b/apps/sim/providers/baseten/index.ts
index 5b00cecfcf0..79463252954 100644
--- a/apps/sim/providers/baseten/index.ts
+++ b/apps/sim/providers/baseten/index.ts
@@ -92,7 +92,7 @@ export const basetenProvider: ProviderConfig = {
       baseURL: 'https://inference.baseten.co/v1',
     })
 
-    const requestedModel = request.model.replace(/^baseten\//, '')
+    const requestedModel = request.model.replace(/^baseten\//i, '')
 
     logger.info('Preparing Baseten request', {
       model: requestedModel,
diff --git a/apps/sim/providers/bedrock/index.test.ts b/apps/sim/providers/bedrock/index.test.ts
index bf4b2fdecdd..487cc092d2f 100644
--- a/apps/sim/providers/bedrock/index.test.ts
+++ b/apps/sim/providers/bedrock/index.test.ts
@@ -22,6 +22,7 @@ vi.mock('@/providers/bedrock/utils', () => ({
   checkForForcedToolUsage: vi.fn(),
   createReadableStreamFromBedrockStream: vi.fn(),
   generateToolUseId: vi.fn().mockReturnValue('tool-1'),
+  getBedrockBaseModelId: (model: string) => model.replace(/^bedrock\//i, ''),
   getBedrockStreamError: vi.fn().mockReturnValue(null),
   // The mocked inference profile above is a Claude model, which supports it.
   supportsToolResultStatus: vi.fn().mockReturnValue(true),
@@ -35,6 +36,8 @@ vi.mock('@/providers/models', () => ({
   getProviderModels: vi.fn().mockReturnValue([]),
   getProviderDefaultModel: vi.fn().mockReturnValue('us.anthropic.claude-3-5-sonnet-20241022-v2:0'),
   supportsNativeStructuredOutputs: vi.fn().mockReturnValue(false),
+  getModelCapabilities: vi.fn().mockReturnValue({ temperature: { min: 0, max: 1 } }),
+  isKnownModelId: vi.fn().mockReturnValue(true),
 }))
 
 vi.mock('@/providers/utils', () => ({
@@ -64,6 +67,7 @@ import { BedrockRuntimeClient, ConverseCommand } from '@aws-sdk/client-bedrock-r
 import type { StreamingExecution } from '@/executor/types'
 import { bedrockProvider } from '@/providers/bedrock/index'
 import { clearProviderClientCacheForTests } from '@/providers/client-cache'
+import { getModelCapabilities, isKnownModelId } from '@/providers/models'
 import { prepareToolsWithUsageControl } from '@/providers/utils'
 
 describe('bedrockProvider credential handling', () => {
@@ -135,6 +139,34 @@ describe('bedrockProvider credential handling', () => {
     })
   })
 
+  it('omits temperature for catalog models that do not support it', async () => {
+    vi.mocked(getModelCapabilities).mockReturnValueOnce({ maxOutputTokens: 128000 })
+    await bedrockProvider.executeRequest({
+      ...baseRequest,
+      model: 'bedrock/anthropic.claude-opus-5',
+      temperature: 0.7,
+    })
+    expect(ConverseCommand).toHaveBeenCalledWith(expect.objectContaining({ inferenceConfig: {} }))
+  })
+
+  it('preserves explicit temperature for a custom model without catalog capabilities', async () => {
+    vi.mocked(isKnownModelId).mockReturnValueOnce(false)
+    await bedrockProvider.executeRequest({
+      ...baseRequest,
+      model: 'bedrock/MyCustomProfile',
+      temperature: 0.2,
+    })
+    expect(ConverseCommand).toHaveBeenCalledWith(
+      expect.objectContaining({ inferenceConfig: { temperature: 0.2 } })
+    )
+  })
+
+  it('leaves temperature to the service default for a custom model when omitted', async () => {
+    vi.mocked(isKnownModelId).mockReturnValueOnce(false)
+    await bedrockProvider.executeRequest({ ...baseRequest, model: 'bedrock/MyCustomProfile' })
+    expect(ConverseCommand).toHaveBeenCalledWith(expect.objectContaining({ inferenceConfig: {} }))
+  })
+
   it('uses the live loop for streaming tool requests without a caller flag', async () => {
     vi.mocked(prepareToolsWithUsageControl).mockReturnValueOnce({
       tools: [
diff --git a/apps/sim/providers/bedrock/index.ts b/apps/sim/providers/bedrock/index.ts
index 28b9b112d27..944785aca63 100644
--- a/apps/sim/providers/bedrock/index.ts
+++ b/apps/sim/providers/bedrock/index.ts
@@ -26,13 +26,16 @@ import {
   checkForForcedToolUsage,
   createReadableStreamFromBedrockStream,
   generateToolUseId,
+  getBedrockBaseModelId,
   getBedrockInferenceProfileId,
   supportsToolResultStatus,
 } from '@/providers/bedrock/utils'
 import { getCachedProviderClient } from '@/providers/client-cache'
 import {
+  getModelCapabilities,
   getProviderDefaultModel,
   getProviderModels,
+  isKnownModelId,
   supportsNativeStructuredOutputs,
 } from '@/providers/models'
 import { executeProviderTool } from '@/providers/runtime-context'
@@ -376,8 +379,15 @@ export const bedrockProvider: ProviderConfig = {
 
     const systemPromptWithSchema = systemContent
 
-    const inferenceConfig: { temperature: number; maxTokens?: number } = {
-      temperature: Number.parseFloat(String(request.temperature ?? 0.7)),
+    const canonicalModelId = `bedrock/${getBedrockBaseModelId(request.model)}`
+    const knownModel = isKnownModelId(canonicalModelId)
+    const modelCapabilities = getModelCapabilities(canonicalModelId)
+    const inferenceConfig: { temperature?: number; maxTokens?: number } = {}
+    if (
+      (knownModel && modelCapabilities?.temperature) ||
+      (!knownModel && request.temperature != null)
+    ) {
+      inferenceConfig.temperature = Number.parseFloat(String(request.temperature ?? 0.7))
     }
     if (request.maxTokens != null) {
       inferenceConfig.maxTokens = Number.parseInt(String(request.maxTokens))
diff --git a/apps/sim/providers/bedrock/streaming-tool-loop.test.ts b/apps/sim/providers/bedrock/streaming-tool-loop.test.ts
index ebb828e1dea..cbc4e73eb00 100644
--- a/apps/sim/providers/bedrock/streaming-tool-loop.test.ts
+++ b/apps/sim/providers/bedrock/streaming-tool-loop.test.ts
@@ -1,6 +1,7 @@
 /**
  * @vitest-environment node
  */
+import type { ConverseStreamCommand } from '@aws-sdk/client-bedrock-runtime'
 import { beforeEach, describe, expect, it, vi } from 'vitest'
 import { createBedrockStreamingToolLoopStream } from '@/providers/bedrock/streaming-tool-loop'
 import type { AgentStreamEvent } from '@/providers/stream-events'
@@ -55,18 +56,51 @@ describe('createBedrockStreamingToolLoopStream', () => {
     })
   })
 
-  it('emits tool_call_start/end and final text; no invented thinking', async () => {
+  it('preserves signed and redacted reasoning across tool turns without exposing it', async () => {
     const turns = [
       (async function* () {
         yield {
-          contentBlockStart: {
+          contentBlockDelta: {
             contentBlockIndex: 0,
-            start: { toolUse: { toolUseId: 'tooluse_1', name: 'http_request' } },
+            delta: { reasoningContent: { text: 'Check the endpoint.' } },
+          },
+        }
+        yield {
+          contentBlockDelta: {
+            contentBlockIndex: 0,
+            delta: { reasoningContent: { signature: 'signature-' } },
           },
         }
         yield {
           contentBlockDelta: {
             contentBlockIndex: 0,
+            delta: { reasoningContent: { signature: 'value' } },
+          },
+        }
+        yield {
+          contentBlockDelta: { contentBlockIndex: 1, delta: { text: 'Checking now.' } },
+        }
+        yield {
+          contentBlockDelta: {
+            contentBlockIndex: 2,
+            delta: { reasoningContent: { redactedContent: new Uint8Array([1, 2]) } },
+          },
+        }
+        yield {
+          contentBlockDelta: {
+            contentBlockIndex: 2,
+            delta: { reasoningContent: { redactedContent: new Uint8Array([3]) } },
+          },
+        }
+        yield {
+          contentBlockStart: {
+            contentBlockIndex: 3,
+            start: { toolUse: { toolUseId: 'tooluse_1', name: 'http_request' } },
+          },
+        }
+        yield {
+          contentBlockDelta: {
+            contentBlockIndex: 3,
             delta: { toolUse: { input: '{"url":"https://example.com"}' } },
           },
         }
@@ -91,7 +125,7 @@ describe('createBedrockStreamingToolLoopStream', () => {
 
     let turnIdx = 0
     const client = {
-      send: vi.fn(async () => ({ stream: turns[turnIdx++] })),
+      send: vi.fn(async (_command: ConverseStreamCommand) => ({ stream: turns[turnIdx++] })),
     }
 
     const onComplete = vi.fn()
@@ -128,6 +162,26 @@ describe('createBedrockStreamingToolLoopStream', () => {
 
     const events = await collectEvents(stream)
 
+    expect(client.send.mock.calls[1][0].input.messages?.[1]).toEqual({
+      role: 'assistant',
+      content: [
+        {
+          reasoningContent: {
+            reasoningText: { text: 'Check the endpoint.', signature: 'signature-value' },
+          },
+        },
+        { text: 'Checking now.' },
+        { reasoningContent: { redactedContent: new Uint8Array([1, 2, 3]) } },
+        {
+          toolUse: {
+            toolUseId: 'tooluse_1',
+            name: 'http_request',
+            input: { url: 'https://example.com' },
+          },
+        },
+      ],
+    })
+
     expect(events.some((e) => e.type === 'thinking_delta')).toBe(false)
     expect(events.filter((e) => e.type === 'tool_call_start')).toEqual([
       { type: 'tool_call_start', id: 'tooluse_1', name: 'http_request' },
@@ -141,7 +195,7 @@ describe('createBedrockStreamingToolLoopStream', () => {
         .filter((e) => e.type === 'text_delta' && e.turn === 'pending')
         .map((e) => e.text)
         .join('')
-    ).toBe('Request completed.')
+    ).toBe('Checking now.Request completed.')
     expect(events.filter((e) => e.type === 'turn_end').map((e) => e.turn)).toEqual([
       'intermediate',
       'final',
diff --git a/apps/sim/providers/bedrock/streaming-tool-loop.ts b/apps/sim/providers/bedrock/streaming-tool-loop.ts
index 0ecd6c276cf..13b3d8646e2 100644
--- a/apps/sim/providers/bedrock/streaming-tool-loop.ts
+++ b/apps/sim/providers/bedrock/streaming-tool-loop.ts
@@ -1,8 +1,8 @@
 /**
  * Live Bedrock ConverseStream tool loop.
  *
- * Capability-honest: text + tool_call_start/end only — Sim does not request
- * Bedrock reasoning, so no thinking is invented. Text emits live as `pending`
+ * Text + tool_call_start/end events, with provider reasoning preserved for
+ * subsequent model requests. Text emits live as `pending`
  * deltas and a `turn_end` event classifies each turn, so the pump projects
  * only final-turn text to the answer channel. Abort → cancelled.
  */
@@ -47,7 +47,7 @@ export interface CreateBedrockStreamingToolLoopStreamOptions {
   request: ProviderRequest
   messages: BedrockMessage[]
   system?: SystemContentBlock[]
-  inferenceConfig: { temperature: number; maxTokens?: number }
+  inferenceConfig: { temperature?: number; maxTokens?: number }
   bedrockTools: Tool[]
   toolChoice: ToolConfiguration['toolChoice']
   logger: Logger
@@ -62,6 +62,8 @@ interface AssembledToolUse {
   inputJson: string
 }
 
+type DrainedContentBlock = ContentBlock | { pendingToolUseId: string }
+
 type ToolUseInput = NonNullable<ToolUseBlock['input']>
 
 function parseToolInput(inputJson: string): Record<string, ToolUseInput> {
@@ -84,12 +86,18 @@ async function drainBedrockTurn(
 ): Promise<{
   text: string
   toolUses: AssembledToolUse[]
+  content: DrainedContentBlock[]
   inputTokens: number
   outputTokens: number
   stopReason?: string
 }> {
   let text = ''
   const toolsByIndex = new Map<number, AssembledToolUse>()
+  const textByIndex = new Map<number, string>()
+  const reasoningByIndex = new Map<
+    number,
+    { text: string; signature: string; redacted: Uint8Array[] }
+  >()
   let currentIndex: number | undefined
   let inputTokens = 0
   let outputTokens = 0
@@ -119,8 +127,23 @@ async function drainBedrockTurn(
     if (event.contentBlockDelta) {
       const idx = event.contentBlockDelta.contentBlockIndex ?? currentIndex
       const delta = event.contentBlockDelta.delta
+      if (delta?.reasoningContent && typeof idx === 'number') {
+        let reasoning = reasoningByIndex.get(idx)
+        if (!reasoning) {
+          reasoning = { text: '', signature: '', redacted: [] }
+          reasoningByIndex.set(idx, reasoning)
+        }
+        reasoning.text += delta.reasoningContent.text ?? ''
+        reasoning.signature += delta.reasoningContent.signature ?? ''
+        if (delta.reasoningContent.redactedContent) {
+          reasoning.redacted.push(delta.reasoningContent.redactedContent)
+        }
+      }
       if (delta?.text) {
         text += delta.text
+        if (typeof idx === 'number') {
+          textByIndex.set(idx, (textByIndex.get(idx) ?? '') + delta.text)
+        }
         // Live pending text: sinks render it now; the pump projects it to the
         // answer only when this turn's turn_end says 'final'.
         controller.enqueue({ type: 'text_delta', text: delta.text, turn: 'pending' })
@@ -145,9 +168,39 @@ async function drainBedrockTurn(
     }
   }
 
+  const contentByIndex = new Map<number, DrainedContentBlock>()
+  for (const [index, blockText] of textByIndex) {
+    if (blockText.trim()) contentByIndex.set(index, { text: blockText })
+  }
+  for (const [index, tool] of toolsByIndex) {
+    contentByIndex.set(index, { pendingToolUseId: tool.toolUseId })
+  }
+  for (const [index, reasoning] of reasoningByIndex) {
+    if (reasoning.redacted.length > 0) {
+      const redactedContent = new Uint8Array(
+        reasoning.redacted.reduce((size, chunk) => size + chunk.length, 0)
+      )
+      let offset = 0
+      for (const chunk of reasoning.redacted) {
+        redactedContent.set(chunk, offset)
+        offset += chunk.length
+      }
+      contentByIndex.set(index, { reasoningContent: { redactedContent } })
+    } else {
+      contentByIndex.set(index, {
+        reasoningContent: {
+          reasoningText: { text: reasoning.text, signature: reasoning.signature },
+        },
+      })
+    }
+  }
+
   return {
     text,
     toolUses: [...toolsByIndex.values()],
+    content: [...contentByIndex.entries()]
+      .sort(([left], [right]) => left - right)
+      .map(([, block]) => block),
     inputTokens,
     outputTokens,
     stopReason,
@@ -499,21 +552,17 @@ export function createBedrockStreamingToolLoopStream(
 
           toolsTime += Date.now() - toolsStartTime
 
-          const assistantContent: ContentBlock[] = [
-            // Bedrock rejects a blank text block, and a model can emit only
-            // whitespace before a tool call.
-            ...(drained.text.trim() ? [{ text: drained.text }] : []),
-            ...assembledToolUses.map((toolUse) => ({
-              toolUse: {
-                toolUseId: toolUse.toolUseId,
-                name: toolUse.name,
-                input: toolUse.input,
-              },
-            })),
-          ]
+          const toolUsesById = new Map(
+            assembledToolUses.map((toolUse) => [toolUse.toolUseId, toolUse])
+          )
           currentMessages.push({
             role: 'assistant' as ConversationRole,
-            content: assistantContent,
+            content: drained.content.map((block) => {
+              if (!('pendingToolUseId' in block)) return block
+              const toolUse = toolUsesById.get(block.pendingToolUseId)
+              if (!toolUse) throw new Error('Missing assembled Bedrock tool use')
+              return { toolUse }
+            }),
           })
 
           const toolResultContent: ContentBlock[] = []
diff --git a/apps/sim/providers/bedrock/utils.test.ts b/apps/sim/providers/bedrock/utils.test.ts
index 421e46bcff4..851a84991aa 100644
--- a/apps/sim/providers/bedrock/utils.test.ts
+++ b/apps/sim/providers/bedrock/utils.test.ts
@@ -26,6 +26,34 @@ describe('getBedrockInferenceProfileId', () => {
     )
   })
 
+  it.each([
+    'arn:aws:bedrock:us-east-1:123456789012:application-inference-profile/MyProfile',
+    'arn:aws:bedrock:us-east-1:123456789012:custom-model-deployment/MyDeployment',
+    'MyCustomProfile',
+    'future.vendor-model-v1:0',
+    'global.anthropic.claude-opus-5',
+  ])('preserves caller-supplied model ID %s', (model) => {
+    expect(getBedrockInferenceProfileId(`BEDROCK/${model}`, 'us-east-1')).toBe(model)
+  })
+
+  it.concurrent('uses only published geographic prefixes for new catalog models', () => {
+    expect(getBedrockInferenceProfileId('bedrock/anthropic.claude-opus-5', 'us-east-1')).toBe(
+      'us.anthropic.claude-opus-5'
+    )
+    expect(getBedrockInferenceProfileId('bedrock/anthropic.claude-opus-5', 'ap-southeast-2')).toBe(
+      'au.anthropic.claude-opus-5'
+    )
+    expect(getBedrockInferenceProfileId('bedrock/openai.gpt-5.6-sol', 'us-east-1')).toBe(
+      'us.openai.gpt-5.6-sol'
+    )
+    expect(() => getBedrockInferenceProfileId('bedrock/openai.gpt-5.6-sol', 'eu-west-1')).toThrow(
+      'Supply an explicit bedrock/global.'
+    )
+    expect(getBedrockInferenceProfileId('bedrock/openai.gpt-oss-120b-1:0', 'us-east-1')).toBe(
+      'openai.gpt-oss-120b-1:0'
+    )
+  })
+
   it.concurrent('returns the bare model ID for models without geo profile support', () => {
     expect(
       getBedrockInferenceProfileId('bedrock/mistral.mistral-large-3-675b-instruct', 'us-east-1')
diff --git a/apps/sim/providers/bedrock/utils.ts b/apps/sim/providers/bedrock/utils.ts
index a8e837b1142..b84cba8c9a0 100644
--- a/apps/sim/providers/bedrock/utils.ts
+++ b/apps/sim/providers/bedrock/utils.ts
@@ -122,22 +122,49 @@ export function generateToolUseId(toolName: string): string {
 }
 
 /**
- * Models whose AWS model cards state geo/cross-region inference profiles are
- * not supported ("Geo inference ID: Not supported"). These must be invoked
- * with the bare in-region model ID — prefixing them with a geo profile
- * (e.g. us.mistral...) produces an invalid model identifier.
+ * Catalog models with documented geographic inference profiles. Unknown model
+ * IDs and caller-supplied inference profile IDs/ARNs must pass through unchanged.
  */
-const GEO_PROFILE_UNSUPPORTED_MODEL_IDS = new Set([
-  'mistral.mistral-large-3-675b-instruct',
-  'mistral.mistral-large-2407-v1:0',
-  'mistral.magistral-small-2509',
-  'mistral.ministral-3-14b-instruct',
-  'mistral.ministral-3-8b-instruct',
-  'mistral.ministral-3-3b-instruct',
-  'mistral.mixtral-8x7b-instruct-v0:1',
-  'amazon.titan-text-premier-v1:0',
-  'cohere.command-r-v1:0',
-  'cohere.command-r-plus-v1:0',
+const GEO_PROFILE_MODEL_IDS = new Set([
+  'anthropic.claude-opus-4-5-20251101-v1:0',
+  'anthropic.claude-sonnet-4-5-20250929-v1:0',
+  'anthropic.claude-haiku-4-5-20251001-v1:0',
+  'anthropic.claude-opus-4-1-20250805-v1:0',
+  'amazon.nova-2-lite-v1:0',
+  'amazon.nova-premier-v1:0',
+  'amazon.nova-pro-v1:0',
+  'amazon.nova-lite-v1:0',
+  'amazon.nova-micro-v1:0',
+  'meta.llama4-maverick-17b-instruct-v1:0',
+  'meta.llama4-scout-17b-instruct-v1:0',
+  'meta.llama3-3-70b-instruct-v1:0',
+  'meta.llama3-2-90b-instruct-v1:0',
+  'meta.llama3-2-11b-instruct-v1:0',
+  'meta.llama3-2-3b-instruct-v1:0',
+  'meta.llama3-2-1b-instruct-v1:0',
+  'meta.llama3-1-405b-instruct-v1:0',
+  'meta.llama3-1-70b-instruct-v1:0',
+  'meta.llama3-1-8b-instruct-v1:0',
+  'mistral.pixtral-large-2502-v1:0',
+])
+
+/** Current Claude profiles use AU rather than the older APAC geography. */
+const CLAUDE_GEO_PROFILE_MODEL_IDS = new Set([
+  'anthropic.claude-opus-5',
+  'anthropic.claude-sonnet-5',
+  'anthropic.claude-opus-4-8',
+  'anthropic.claude-opus-4-7',
+  'anthropic.claude-opus-4-6-v1',
+  'anthropic.claude-sonnet-4-6',
+])
+
+/** These models currently publish US and global inference profiles only. */
+const US_GEO_PROFILE_MODEL_IDS = new Set([
+  'anthropic.claude-fable-5',
+  'openai.gpt-6-astra',
+  'openai.gpt-5.6-sol',
+  'openai.gpt-5.6-terra',
+  'openai.gpt-5.6-luna',
 ])
 
 /** Cross-region inference profile prefixes Bedrock prepends to a base model ID. */
@@ -147,8 +174,8 @@ const GEO_PROFILE_PREFIX_PATTERN = /^(us-gov|us|eu|apac|au|ca|jp|global)\./
  * Strips Sim's `bedrock/` namespace and any cross-region inference prefix,
  * leaving the bare `<vendor>.<model>` ID that capability checks key off.
  */
-function getBedrockBaseModelId(modelId: string): string {
-  const withoutNamespace = modelId.startsWith('bedrock/') ? modelId.slice(8) : modelId
+export function getBedrockBaseModelId(modelId: string): string {
+  const withoutNamespace = modelId.replace(/^bedrock\//i, '')
   return withoutNamespace.replace(GEO_PROFILE_PREFIX_PATTERN, '')
 }
 
@@ -177,13 +204,33 @@ export function supportsToolResultStatus(modelId: string): boolean {
  * @returns The inference profile ID (e.g., "us.anthropic.claude-sonnet-4-5-20250929-v1:0")
  */
 export function getBedrockInferenceProfileId(modelId: string, region: string): string {
-  const baseModelId = modelId.startsWith('bedrock/') ? modelId.slice(8) : modelId
+  const baseModelId = modelId.replace(/^bedrock\//i, '')
 
   if (GEO_PROFILE_PREFIX_PATTERN.test(baseModelId)) {
     return baseModelId
   }
 
-  if (GEO_PROFILE_UNSUPPORTED_MODEL_IDS.has(baseModelId)) {
+  if (CLAUDE_GEO_PROFILE_MODEL_IDS.has(baseModelId)) {
+    if ((region.startsWith('us-') && !region.startsWith('us-gov-')) || region.startsWith('ca-')) {
+      return `us.${baseModelId}`
+    }
+    if (region.startsWith('eu-')) return `eu.${baseModelId}`
+    if (region === 'ap-southeast-2' || region === 'ap-southeast-4') return `au.${baseModelId}`
+    throw new Error(
+      `No geographic inference profile is configured for ${baseModelId} in ${region}. ` +
+        'Supply an explicit bedrock/global. model ID or an inference profile ARN.'
+    )
+  }
+
+  if (US_GEO_PROFILE_MODEL_IDS.has(baseModelId)) {
+    if (region.startsWith('us-') && !region.startsWith('us-gov-')) return `us.${baseModelId}`
+    throw new Error(
+      `${baseModelId} only has a US geographic inference profile. ` +
+        'Supply an explicit bedrock/global. model ID to use global inference.'
+    )
+  }
+
+  if (!GEO_PROFILE_MODEL_IDS.has(baseModelId)) {
     return baseModelId
   }
 
diff --git a/apps/sim/providers/cerebras/index.ts b/apps/sim/providers/cerebras/index.ts
index bb79fcfa807..14613a0b7bb 100644
--- a/apps/sim/providers/cerebras/index.ts
+++ b/apps/sim/providers/cerebras/index.ts
@@ -81,11 +81,14 @@ export const cerebrasProvider: ProviderConfig = {
         : undefined
 
       const payload: any = {
-        model: request.model.replace('cerebras/', ''),
+        model: request.model.replace(/^cerebras\//i, ''),
         messages: formattedMessages,
       }
       if (request.temperature !== undefined) payload.temperature = request.temperature
       if (request.maxTokens != null) payload.max_completion_tokens = request.maxTokens
+      if (request.reasoningEffort && request.reasoningEffort !== 'auto') {
+        payload.reasoning_effort = request.reasoningEffort
+      }
       if (request.responseFormat) {
         payload.response_format = {
           type: 'json_schema',
diff --git a/apps/sim/providers/deepseek/index.test.ts b/apps/sim/providers/deepseek/index.test.ts
index 174123c7ffa..c1c363f38a1 100644
--- a/apps/sim/providers/deepseek/index.test.ts
+++ b/apps/sim/providers/deepseek/index.test.ts
@@ -108,6 +108,20 @@ describe('deepseekProvider thinking payload', () => {
     expect(payload.thinking).toBeUndefined()
   })
 
+  it.each([
+    ['low', 'low'],
+    ['minimal', 'low'],
+    ['medium', 'high'],
+    ['xhigh', 'high'],
+    ['max', 'max'],
+  ] as const)('maps Flash reasoning effort %s to %s', async (reasoningEffort, expected) => {
+    await deepseekProvider.executeRequest(request({ model: 'deepseek-flash', reasoningEffort }))
+    expect(mockCreate.mock.calls[0][0]).toMatchObject({
+      model: 'deepseek-flash',
+      reasoning_effort: expected,
+    })
+  })
+
   it('selects the live tool loop without a caller flag', async () => {
     mockPrepareToolsWithUsageControl.mockReturnValue({
       tools: [
diff --git a/apps/sim/providers/deepseek/index.ts b/apps/sim/providers/deepseek/index.ts
index 2897d8e9ab0..7dd067f5a26 100644
--- a/apps/sim/providers/deepseek/index.ts
+++ b/apps/sim/providers/deepseek/index.ts
@@ -105,9 +105,9 @@ export const deepseekProvider: ProviderConfig = {
       }
       if (request.reasoningEffort && !['auto', 'none'].includes(request.reasoningEffort)) {
         payload.reasoning_effort =
-          request.reasoningEffort === 'xhigh'
-            ? 'max'
-            : request.reasoningEffort === 'low' || request.reasoningEffort === 'medium'
+          request.reasoningEffort === 'minimal'
+            ? 'low'
+            : request.reasoningEffort === 'xhigh' || request.reasoningEffort === 'medium'
               ? 'high'
               : request.reasoningEffort
       }
diff --git a/apps/sim/providers/fireworks/index.test.ts b/apps/sim/providers/fireworks/index.test.ts
index 9dcd6d9cfe5..8a5ef735ad3 100644
--- a/apps/sim/providers/fireworks/index.test.ts
+++ b/apps/sim/providers/fireworks/index.test.ts
@@ -127,6 +127,18 @@ describe('fireworksProvider', () => {
     apiKey: 'fw-test-key',
   }
 
+  it('preserves custom deployment paths when stripping an uppercase namespace', async () => {
+    mockCreate.mockResolvedValueOnce(textResponse('ok'))
+    await fireworksProvider.executeRequest({
+      ...baseRequest,
+      model: 'FIREWORKS/accounts/Example/models/CustomModel',
+    })
+    expect(mockResolveFireworksWireModel).toHaveBeenCalledWith(
+      'accounts/Example/models/CustomModel'
+    )
+    expect(callBody(0).model).toBe('accounts/Example/models/CustomModel')
+  })
+
   it('throws when the API key is missing', async () => {
     await expect(
       fireworksProvider.executeRequest({ ...baseRequest, apiKey: undefined })
diff --git a/apps/sim/providers/fireworks/index.ts b/apps/sim/providers/fireworks/index.ts
index 5ebb31606cd..f82370e02d8 100644
--- a/apps/sim/providers/fireworks/index.ts
+++ b/apps/sim/providers/fireworks/index.ts
@@ -93,7 +93,7 @@ export const fireworksProvider: ProviderConfig = {
       baseURL: 'https://api.fireworks.ai/inference/v1',
     })
 
-    const requestedModel = resolveFireworksWireModel(request.model.replace(/^fireworks\//, ''))
+    const requestedModel = resolveFireworksWireModel(request.model.replace(/^fireworks\//i, ''))
 
     logger.info('Preparing Fireworks request', {
       model: requestedModel,
diff --git a/apps/sim/providers/gemini/core.request.test.ts b/apps/sim/providers/gemini/core.request.test.ts
new file mode 100644
index 00000000000..1d813deb3a5
--- /dev/null
+++ b/apps/sim/providers/gemini/core.request.test.ts
@@ -0,0 +1,172 @@
+/**
+ * @vitest-environment node
+ */
+import type { GenerateContentParameters, GenerateContentResponse } from '@google/genai'
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+import type { StreamingExecution } from '@/executor/types'
+import { executeGeminiRequest } from '@/providers/gemini/core'
+import type { ProviderRequest, ProviderResponse } from '@/providers/types'
+
+const { mockExecuteTool } = vi.hoisted(() => ({ mockExecuteTool: vi.fn() }))
+
+vi.mock('@/tools', () => ({ executeTool: mockExecuteTool }))
+vi.mock('@/providers', () => ({ MAX_TOOL_ITERATIONS: 5 }))
+
+function textTurn(): GenerateContentResponse {
+  return {
+    candidates: [{ content: { role: 'model', parts: [{ text: 'answer' }] }, finishReason: 'STOP' }],
+    usageMetadata: {
+      promptTokenCount: 1000,
+      cachedContentTokenCount: 200,
+      candidatesTokenCount: 100,
+      totalTokenCount: 1100,
+    },
+  } as GenerateContentResponse
+}
+
+async function run(
+  model: string,
+  generateContent: ReturnType<typeof vi.fn>,
+  overrides: Partial<ProviderRequest> = {}
+) {
+  return (await executeGeminiRequest({
+    ai: { models: { generateContent, generateContentStream: vi.fn() } } as never,
+    model: model.replace(/^vertex\//i, ''),
+    providerType: 'vertex',
+    request: {
+      model,
+      apiKey: 'test-key',
+      messages: [{ role: 'user', content: 'Look this up' }],
+      temperature: 0.5,
+      thinkingLevel: 'medium',
+      ...overrides,
+    },
+  })) as ProviderResponse
+}
+
+describe('Vertex Gemini request compatibility', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+    mockExecuteTool.mockResolvedValue({ success: true, output: { value: 'tool result' } })
+  })
+
+  it.each([
+    'vertex/gemini-3.8-flash',
+    'vertex/gemini-3.7-flash',
+    'vertex/gemini-3.6-flash',
+    'vertex/gemini-3.5-flash-lite',
+  ])('uses thinking levels and omits ignored temperature for %s', async (model) => {
+    const generateContent = vi.fn().mockResolvedValue(textTurn())
+    await run(model, generateContent)
+
+    const params = generateContent.mock.calls[0][0] as GenerateContentParameters
+    expect(params.model).toBe(model.replace('vertex/', ''))
+    expect(params.config).not.toHaveProperty('temperature')
+    expect(params.config?.thinkingConfig).toEqual({
+      thinkingLevel: 'MEDIUM',
+      includeThoughts: false,
+    })
+  })
+
+  it.each(['vertex/gemini-2.5-flash', 'vertex/Custom-Gemini-Model'])(
+    'preserves temperature for supported or uncataloged %s',
+    async (model) => {
+      const generateContent = vi.fn().mockResolvedValue(textTurn())
+      await run(model, generateContent)
+
+      expect(generateContent.mock.calls[0][0].config.temperature).toBe(0.5)
+    }
+  )
+
+  it('prices Vertex-only catalog entries using the namespaced model ID', async () => {
+    const generateContent = vi.fn().mockResolvedValue(textTurn())
+    const result = await run('vertex/gemini-3.8-flash', generateContent)
+
+    expect(result.cost?.input).toBeCloseTo((800 * 0.75 + 200 * 0.075) / 1e6, 10)
+    expect(result.cost?.output).toBeCloseTo((100 * 3.75) / 1e6, 10)
+  })
+
+  it.each([false, true])(
+    'retains Vertex pricing when streaming with tools enabled: %s',
+    async (withTools) => {
+      const generateContentStream = vi.fn().mockImplementation(async function* () {
+        yield textTurn()
+      })
+      const result = (await executeGeminiRequest({
+        ai: { models: { generateContentStream } } as never,
+        model: 'gemini-3.8-flash',
+        providerType: 'vertex',
+        request: {
+          model: 'vertex/gemini-3.8-flash',
+          messages: [{ role: 'user', content: 'Hello' }],
+          stream: true,
+          ...(withTools
+            ? {
+                tools: [
+                  {
+                    id: 'lookup',
+                    name: 'lookup',
+                    description: 'Look up a query',
+                    parameters: { type: 'object', properties: {}, required: [] },
+                  },
+                ],
+              }
+            : {}),
+        },
+      })) as StreamingExecution
+      const reader = result.stream.getReader()
+      while (!(await reader.read()).done) {}
+      reader.releaseLock()
+
+      expect(result.execution.output.cost?.input).toBeCloseTo((800 * 0.75 + 200 * 0.075) / 1e6, 10)
+      expect(result.execution.output.cost?.output).toBeCloseTo((100 * 3.75) / 1e6, 10)
+    }
+  )
+
+  it('echoes IDs and thought signatures for parallel same-name function calls', async () => {
+    const functionCalls = [
+      { id: 'call-first', name: 'lookup', args: { query: 'first' } },
+      { id: 'call-second', name: 'lookup', args: { query: 'second' } },
+    ]
+    const parts = functionCalls.map((functionCall, index) => ({
+      functionCall,
+      thoughtSignature: `signature-${index}`,
+    }))
+    const generateContent = vi
+      .fn()
+      .mockResolvedValueOnce({
+        candidates: [{ content: { role: 'model', parts }, finishReason: 'STOP' }],
+        functionCalls,
+        usageMetadata: { promptTokenCount: 10, candidatesTokenCount: 20, totalTokenCount: 30 },
+      })
+      .mockResolvedValueOnce(textTurn())
+
+    await run('vertex/gemini-3.8-flash', generateContent, {
+      tools: [
+        {
+          id: 'lookup',
+          name: 'lookup',
+          description: 'Look up a query',
+          parameters: {
+            type: 'object',
+            properties: { query: { type: 'string' } },
+            required: ['query'],
+          },
+        },
+      ],
+    })
+
+    expect(generateContent).toHaveBeenCalledTimes(2)
+    const followUp = generateContent.mock.calls[1][0] as GenerateContentParameters
+    expect(followUp.contents).toEqual([
+      expect.objectContaining({ role: 'user' }),
+      { role: 'model', parts },
+      {
+        role: 'user',
+        parts: functionCalls.map(({ id, name }) => ({
+          functionResponse: { id, name, response: { value: 'tool result' } },
+        })),
+      },
+    ])
+  })
+})
diff --git a/apps/sim/providers/gemini/core.ts b/apps/sim/providers/gemini/core.ts
index 175b71d96b7..d47e4c06cfa 100644
--- a/apps/sim/providers/gemini/core.ts
+++ b/apps/sim/providers/gemini/core.ts
@@ -31,6 +31,7 @@ import {
   mapToThinkingLevel,
   supportsDisablingGemini25Thinking,
 } from '@/providers/google/utils'
+import { getModelCapabilities, isKnownModelId } from '@/providers/models'
 import { executeProviderTool } from '@/providers/runtime-context'
 import { createSettledAgentEventStream } from '@/providers/stream-events'
 import { createStreamingExecution } from '@/providers/streaming-execution'
@@ -215,6 +216,7 @@ async function executeToolCallsBatch(
     functionResponse: {
       name: r.toolName,
       response: 'modelResultContent' in r ? r.modelResultContent : r.resultContent,
+      ...(r.part.functionCall?.id ? { id: r.part.functionCall.id } : {}),
     },
   }))
 
@@ -966,7 +968,10 @@ export async function executeGeminiRequest(
     if (request.abortSignal) {
       geminiConfig.abortSignal = request.abortSignal
     }
-    if (request.temperature !== undefined) {
+    if (
+      request.temperature !== undefined &&
+      (!isKnownModelId(request.model) || getModelCapabilities(request.model)?.temperature)
+    ) {
       geminiConfig.temperature = request.temperature
     }
     if (request.maxTokens != null) {
@@ -1148,7 +1153,7 @@ export async function executeGeminiRequest(
 
           streamingResult.execution.output.content = content
           streamingResult.execution.output.tokens = { ...split, total: usage.totalTokenCount }
-          streamingResult.execution.output.cost = priceGeminiTokens(model, split)
+          streamingResult.execution.output.cost = priceGeminiTokens(request.model, split)
 
           if (thinking) {
             const segment = streamingResult.execution.output.providerTiming?.timeSegments?.[0]
@@ -1192,11 +1197,11 @@ export async function executeGeminiRequest(
       initialUsage,
       firstResponseTime,
       initialCallTime,
-      model,
+      request.model,
       toolConfig
     )
     enrichLastModelSegmentFromGeminiResponse(state.timeSegments, response, {
-      model,
+      model: request.model,
     })
     const forcedTools = preparedTools?.forcedTools ?? []
 
@@ -1225,12 +1230,12 @@ export async function executeGeminiRequest(
       const finalState = updateStateWithResponse(
         currentState,
         finalResponse,
-        model,
+        request.model,
         finalStartTime,
         Date.now()
       )
       enrichLastModelSegmentFromGeminiResponse(finalState.timeSegments, finalResponse, {
-        model,
+        model: request.model,
       })
       return { state: finalState, response: finalResponse }
     }
@@ -1324,9 +1329,15 @@ export async function executeGeminiRequest(
           contents: state.contents,
           config: nextConfig,
         })
-        state = updateStateWithResponse(state, nextResponse, model, nextModelStartTime, Date.now())
+        state = updateStateWithResponse(
+          state,
+          nextResponse,
+          request.model,
+          nextModelStartTime,
+          Date.now()
+        )
         enrichLastModelSegmentFromGeminiResponse(state.timeSegments, nextResponse, {
-          model,
+          model: request.model,
         })
         currentResponse = nextResponse
 
diff --git a/apps/sim/providers/gemini/streaming-tool-loop.ts b/apps/sim/providers/gemini/streaming-tool-loop.ts
index 25167caeb19..646f19acfca 100644
--- a/apps/sim/providers/gemini/streaming-tool-loop.ts
+++ b/apps/sim/providers/gemini/streaming-tool-loop.ts
@@ -339,7 +339,7 @@ export function createGeminiStreamingToolLoopStream(
             tokens.cacheRead += split.cacheRead
             tokens.total += drained.usage.totalTokenCount
 
-            const turnCost = priceGeminiTokens(model, split)
+            const turnCost = priceGeminiTokens(request.model, split)
             costInput += turnCost.input
             costOutput += turnCost.output
             costTotal += turnCost.total
diff --git a/apps/sim/providers/google/utils.test.ts b/apps/sim/providers/google/utils.test.ts
index ead73247543..12f21ffdc2e 100644
--- a/apps/sim/providers/google/utils.test.ts
+++ b/apps/sim/providers/google/utils.test.ts
@@ -280,6 +280,10 @@ describe('convertToGeminiFormat', () => {
 
       const result = convertToGeminiFormat(request)
 
+      expect(result.contents[1].parts?.[0].functionCall).toMatchObject({
+        id: 'call_123',
+        name: 'get_weather',
+      })
       const toolResponseContent = result.contents.find(
         (c) => c.parts?.[0] && 'functionResponse' in c.parts[0]
       )
diff --git a/apps/sim/providers/google/utils.ts b/apps/sim/providers/google/utils.ts
index 63822da75d0..79aa652fb74 100644
--- a/apps/sim/providers/google/utils.ts
+++ b/apps/sim/providers/google/utils.ts
@@ -162,6 +162,7 @@ export function convertToGeminiFormat(
         if (message.role === 'assistant' && message.tool_calls?.length) {
           const functionCalls = message.tool_calls.map((toolCall) => ({
             functionCall: {
+              id: toolCall.id,
               name: toolCall.function?.name,
               args: JSON.parse(toolCall.function?.arguments || '{}') as Record<string, unknown>,
             },
diff --git a/apps/sim/providers/groq/index.test.ts b/apps/sim/providers/groq/index.test.ts
index a10fdce4454..b946d249153 100644
--- a/apps/sim/providers/groq/index.test.ts
+++ b/apps/sim/providers/groq/index.test.ts
@@ -140,6 +140,23 @@ describe('groqProvider reasoning payload', () => {
     expect(payload.reasoning_effort).toBe('none')
   })
 
+  it.each(['none', 'low', 'medium', 'high'] as const)(
+    'Qwen 3.8 forwards explicit reasoning effort %s',
+    async (reasoningEffort) => {
+      await groqProvider.executeRequest(
+        request({ model: 'groq/qwen/qwen3.8-27b', reasoningEffort })
+      )
+      const payload = mockCreate.mock.calls[0][0]
+      expect(payload.reasoning_effort).toBe(reasoningEffort)
+      expect(payload.reasoning_format).toBe(reasoningEffort === 'none' ? undefined : 'parsed')
+    }
+  )
+
+  it('strips only the leading routing prefix and preserves custom model case', async () => {
+    await groqProvider.executeRequest(request({ model: 'Groq/Custom/Model-A' }))
+    expect(mockCreate.mock.calls[0][0].model).toBe('Custom/Model-A')
+  })
+
   it('selects the live tool loop without a caller flag', async () => {
     mockPrepareToolsWithUsageControl.mockReturnValue({
       tools: [
diff --git a/apps/sim/providers/groq/index.ts b/apps/sim/providers/groq/index.ts
index 6bccdb9479a..4b6c310b96f 100644
--- a/apps/sim/providers/groq/index.ts
+++ b/apps/sim/providers/groq/index.ts
@@ -79,7 +79,7 @@ export const groqProvider: ProviderConfig = {
       : undefined
 
     const payload: any = {
-      model: request.model.replace('groq/', ''),
+      model: request.model.replace(/^groq\//i, ''),
       messages: formattedMessages,
     }
 
@@ -101,6 +101,9 @@ export const groqProvider: ProviderConfig = {
     if (isGptOss && (hasExplicitEffort || hasThinkingLevel)) {
       payload.include_reasoning = true
       payload.reasoning_effort = hasExplicitEffort ? request.reasoningEffort : 'medium'
+    } else if (isQwenReasoning && hasExplicitEffort) {
+      payload.reasoning_effort = request.reasoningEffort
+      if (request.reasoningEffort !== 'none') payload.reasoning_format = 'parsed'
     } else if (isQwenReasoning && hasThinkingLevel) {
       payload.reasoning_format = 'parsed'
     } else if (isQwenReasoning && request.thinkingLevel === 'none') {
diff --git a/apps/sim/providers/index.test.ts b/apps/sim/providers/index.test.ts
index f6f962b71e5..e7219714432 100644
--- a/apps/sim/providers/index.test.ts
+++ b/apps/sim/providers/index.test.ts
@@ -1599,6 +1599,30 @@ describe('executeProviderRequest — model level normalization', () => {
     expect(sentRequest().reasoningEffort).toBe('high')
   })
 
+  it.each([
+    ['azure-openai', 'azure/MyDeployment'],
+    ['azure-anthropic', 'azure-anthropic/MyDeployment'],
+    ['bedrock', 'bedrock/custom-inference-profile'],
+    ['vertex', 'vertex/custom-gemini'],
+  ])('preserves tuning levels for a custom %s deployment', async (provider, model) => {
+    await executeProviderRequest(provider, {
+      model,
+      workspaceId: 'ws-1',
+      reasoningEffort: 'high',
+      verbosity: 'low',
+      thinkingLevel: 'high',
+      temperature: 0.7,
+    })
+
+    expect(sentRequest()).toMatchObject({
+      model,
+      reasoningEffort: 'high',
+      verbosity: 'low',
+      thinkingLevel: 'high',
+      temperature: 0.7,
+    })
+  })
+
   it('still drops levels for a dynamic-provider model that does not take them', async () => {
     await executeProviderRequest('ollama', {
       model: 'ollama/llama3',
diff --git a/apps/sim/providers/index.ts b/apps/sim/providers/index.ts
index 01a64387c99..8c7b315f677 100644
--- a/apps/sim/providers/index.ts
+++ b/apps/sim/providers/index.ts
@@ -114,10 +114,6 @@ function sanitizeRequest(request: ProviderRequest): ProviderRequest {
   sanitizedRequest.verbosity = normalizeModelLevel(sanitizedRequest.verbosity)
   sanitizedRequest.thinkingLevel = normalizeModelLevel(sanitizedRequest.thinkingLevel)
 
-  if (model && !supportsTemperature(model)) {
-    sanitizedRequest.temperature = undefined
-  }
-
   /**
    * A model absent from the catalogue is unknown, not known-incapable. The model field is an
    * editable combobox, so a model newer than `models.ts` reaches this point routed by pattern
@@ -128,6 +124,10 @@ function sanitizeRequest(request: ProviderRequest): ProviderRequest {
    */
   const isCatalogued = Boolean(model) && isKnownModelId(model)
 
+  if (model && isCatalogued && !supportsTemperature(model)) {
+    sanitizedRequest.temperature = undefined
+  }
+
   if (model && isCatalogued && !supportsReasoningEffort(model)) {
     sanitizedRequest.reasoningEffort = undefined
   }
diff --git a/apps/sim/providers/kimi/index.ts b/apps/sim/providers/kimi/index.ts
index 0b25f27b5fe..c36a2450116 100644
--- a/apps/sim/providers/kimi/index.ts
+++ b/apps/sim/providers/kimi/index.ts
@@ -48,6 +48,16 @@ const THINKING_TOGGLE_MODELS = new Set(
   )
 )
 
+function buildRequiredToolPayload(
+  tools: OpenAI.Chat.Completions.ChatCompletionTool[],
+  name: string
+) {
+  return {
+    tools: tools.filter((tool) => tool.type === 'function' && tool.function.name === name),
+    tool_choice: 'required' as const,
+  }
+}
+
 function buildResponseFormatPayload(
   responseFormat: NonNullable<ProviderRequest['responseFormat']>
 ) {
@@ -78,8 +88,8 @@ function buildResponseFormatPayload(
  *   rejects the object form whenever thinking is enabled ("tool_choice 'specified' is
  *   incompatible with thinking enabled", verified live). On models with a thinking toggle the
  *   adapter therefore sends `thinking: { type: "disabled" }` for the duration of a forced-tool
- *   request; on always-thinking models (kimi-k3, kimi-k2.7-code) it downgrades the forced
- *   choice to `"auto"` with a warning, mirroring the Z.ai adapter's behavior.
+ *   request. K3 supports `required`, so it forces a named tool by offering only that tool;
+ *   K2.7 Code falls back to `auto` because it supports neither forcing mechanism.
  */
 export const kimiProvider: ProviderConfig = {
   id: 'kimi',
@@ -137,6 +147,9 @@ export const kimiProvider: ProviderConfig = {
       }
 
       if (request.maxTokens != null) payload.max_completion_tokens = request.maxTokens
+      if (request.reasoningEffort && request.reasoningEffort !== 'auto') {
+        payload.reasoning_effort = request.reasoningEffort
+      }
 
       if (
         THINKING_TOGGLE_MODELS.has(request.model) &&
@@ -162,7 +175,12 @@ export const kimiProvider: ProviderConfig = {
           hasActiveTools = true
 
           if (typeof toolChoice === 'object') {
-            if (THINKING_TOGGLE_MODELS.has(request.model)) {
+            if (request.model === 'kimi-k3' && toolChoice.type === 'function') {
+              Object.assign(
+                payload,
+                buildRequiredToolPayload(filteredTools, toolChoice.function.name)
+              )
+            } else if (THINKING_TOGGLE_MODELS.has(request.model)) {
               if (payload.thinking?.type === 'enabled') {
                 logger.warn(
                   'Kimi rejects forced tool_choice while thinking is enabled — disabling thinking for this forced-tool request',
@@ -241,7 +259,8 @@ export const kimiProvider: ProviderConfig = {
       }
 
       const initialCallTime = Date.now()
-      const originalToolChoice = payload.tool_choice
+      const originalToolChoice =
+        request.model === 'kimi-k3' ? preparedTools?.toolChoice : payload.tool_choice
       const forcedTools = preparedTools?.forcedTools || []
       let usedForcedTools: string[] = []
 
@@ -457,22 +476,33 @@ export const kimiProvider: ProviderConfig = {
           const nextPayload = {
             ...payload,
             messages: currentMessages,
+            tools: preparedTools?.tools,
           }
+          let nextToolChoice = nextPayload.tool_choice
 
           if (
             typeof originalToolChoice === 'object' &&
-            hasUsedForcedTool &&
+            (hasUsedForcedTool || request.model === 'kimi-k3') &&
             forcedTools.length > 0
           ) {
             const remainingTools = forcedTools.filter((tool) => !usedForcedTools.includes(tool))
 
             if (remainingTools.length > 0) {
-              nextPayload.tool_choice = {
+              nextToolChoice = {
                 type: 'function',
                 function: { name: remainingTools[0] },
               }
+              if (request.model === 'kimi-k3') {
+                Object.assign(
+                  nextPayload,
+                  buildRequiredToolPayload(preparedTools?.tools || [], remainingTools[0])
+                )
+              } else {
+                nextPayload.tool_choice = nextToolChoice
+              }
               logger.info(`Forcing next tool: ${remainingTools[0]}`)
             } else {
+              nextToolChoice = 'auto'
               nextPayload.tool_choice = 'auto'
               logger.info('All forced tools have been used, switching to auto tool_choice')
             }
@@ -486,10 +516,10 @@ export const kimiProvider: ProviderConfig = {
 
           const toolCallsResponse =
             currentResponse.choices[0]?.message?.tool_calls?.filter(isFunctionToolCall)
-          if (typeof nextPayload.tool_choice === 'object' && toolCallsResponse?.length) {
+          if (typeof nextToolChoice === 'object' && toolCallsResponse?.length) {
             const result = trackForcedToolUsage(
               toolCallsResponse,
-              nextPayload.tool_choice,
+              nextToolChoice,
               logger,
               'openai',
               forcedTools,
diff --git a/apps/sim/providers/litellm/index.test.ts b/apps/sim/providers/litellm/index.test.ts
index 09525507656..7bc9c3253d0 100644
--- a/apps/sim/providers/litellm/index.test.ts
+++ b/apps/sim/providers/litellm/index.test.ts
@@ -140,6 +140,11 @@ describe('litellmProvider.executeRequest', () => {
     mockExecuteTool.mockResolvedValue({ success: true, output: { ok: true } })
   })
 
+  it('preserves a custom proxy model name when stripping an uppercase namespace', async () => {
+    await run({ model: 'LITELLM/Org/CustomModel' })
+    expect(firstPayload().model).toBe('Org/CustomModel')
+  })
+
   it('assembles messages, strips the model prefix, and maps params', async () => {
     const result = await run({
       systemPrompt: 'You are helpful.',
diff --git a/apps/sim/providers/litellm/index.ts b/apps/sim/providers/litellm/index.ts
index 2bdcb7e4a66..aa670263b3e 100644
--- a/apps/sim/providers/litellm/index.ts
+++ b/apps/sim/providers/litellm/index.ts
@@ -142,7 +142,7 @@ export const litellmProvider: ProviderConfig = {
       : undefined
 
     const payload: any = {
-      model: request.model.replace(/^litellm\//, ''),
+      model: request.model.replace(/^litellm\//i, ''),
       messages: formattedMessages,
     }
 
diff --git a/apps/sim/providers/models.test.ts b/apps/sim/providers/models.test.ts
index d6fa3568d05..bcb1d103709 100644
--- a/apps/sim/providers/models.test.ts
+++ b/apps/sim/providers/models.test.ts
@@ -3,6 +3,7 @@
  */
 import { describe, expect, it } from 'vitest'
 import {
+  findProviderFromModel,
   getBaseModelProviders,
   getHostedModels,
   getModelCapabilities,
@@ -11,14 +12,100 @@ import {
   getPromptCachingMinimumTokens,
   getProviderModels,
   getThinkingStreamVisibility,
+  isCustomModelId,
+  isKnownModelId,
   isModelDeprecated,
   orderModelIdsByReleaseDate,
   PROVIDER_DEFINITIONS,
   supportsForcedToolUse,
   updateFireworksModels,
+  updateOllamaModels,
 } from '@/providers/models'
 import { supportsPromptCaching } from '@/providers/utils'
 
+describe('custom cloud model routing', () => {
+  it.each([
+    'ollama',
+    'ollama-cloud',
+    'vllm',
+    'litellm',
+    'openrouter',
+    'fireworks',
+    'together',
+    'baseten',
+  ] as const)(
+    'accepts new model IDs in the %s namespace without accepting an empty ID',
+    (provider) => {
+      expect(findProviderFromModel(`${provider.toUpperCase()}/Org/CustomModel`)).toBe(provider)
+      expect(isKnownModelId(`${provider}/Org/CustomModel`)).toBe(true)
+      expect(isKnownModelId(`${provider}/`)).toBe(false)
+      expect(isKnownModelId(`${provider}/ `)).toBe(false)
+    }
+  )
+
+  it('keeps explicit provider namespaces authoritative over discovered local model names', () => {
+    const originalModels = PROVIDER_DEFINITIONS.ollama.models
+    try {
+      updateOllamaModels([
+        'azure/MyDeployment',
+        'bedrock/CustomModel',
+        'vertex/CustomModel',
+        'openrouter/Org/CustomModel',
+        'groq/Org/CustomModel',
+        'cerebras/CustomModel',
+      ])
+      expect(findProviderFromModel('azure/MyDeployment')).toBe('azure-openai')
+      expect(findProviderFromModel('bedrock/CustomModel')).toBe('bedrock')
+      expect(findProviderFromModel('vertex/CustomModel')).toBe('vertex')
+      expect(findProviderFromModel('openrouter/Org/CustomModel')).toBe('openrouter')
+      expect(findProviderFromModel('groq/Org/CustomModel')).toBe('groq')
+      expect(findProviderFromModel('cerebras/CustomModel')).toBe('cerebras')
+    } finally {
+      PROVIDER_DEFINITIONS.ollama.models = originalModels
+    }
+  })
+
+  it.each([
+    ['azure/MyDeployment', 'azure-openai'],
+    ['AZURE/MyDeployment', 'azure-openai'],
+    ['azure-anthropic/MyDeployment', 'azure-anthropic'],
+    ['bedrock/custom-model:0', 'bedrock'],
+    ['BEDROCK/custom-model:0', 'bedrock'],
+    ['vertex/publishers/google/models/custom-gemini', 'vertex'],
+    ['VERTEX/CustomModel', 'vertex'],
+    ['GROQ/Org/CustomModel', 'groq'],
+    ['CEREBRAS/CustomModel', 'cerebras'],
+    ['NVIDIA/CustomModel', 'nvidia'],
+  ])('routes %s without requiring a catalog entry', (model, provider) => {
+    expect(findProviderFromModel(model)).toBe(provider)
+    expect(isCustomModelId(model)).toBe(true)
+    expect(isKnownModelId(model)).toBe(false)
+    expect(getModelPricing(model)).toBeNull()
+    expect(getHostedModels()).not.toContain(model)
+  })
+
+  it.each([
+    'azure/',
+    'azure/ ',
+    'azure-anthropic/',
+    'bedrock/',
+    'vertex/',
+    'groq/',
+    'cerebras/',
+    'nvidia/',
+    'unknown/model',
+    'gpt-100/model',
+    'mistral/model',
+  ])('does not accept an empty or unrecognized namespace as a custom model: %s', (model) => {
+    expect(isCustomModelId(model)).toBe(false)
+  })
+
+  it('keeps catalog name typos distinct from custom reseller IDs', () => {
+    expect(isCustomModelId('claude-sonnet-4.6')).toBe(false)
+    expect(isCustomModelId('gpt-100-ultra')).toBe(false)
+  })
+})
+
 describe('OpenAI provider definition', () => {
   const openai = PROVIDER_DEFINITIONS.openai
 
@@ -52,6 +139,19 @@ describe('OpenAI provider definition', () => {
   })
 })
 
+describe('direct provider catalog additions', () => {
+  it.each([
+    ['chat-latest', 'openai'],
+    ['gpt-5.3-codex', 'openai'],
+    ['gemini-3.7-flash', 'google'],
+  ])('routes %s through its hosted provider %s', (model, provider) => {
+    expect(findProviderFromModel(model)).toBe(provider)
+    expect(isKnownModelId(model)).toBe(true)
+    expect(getHostedModels()).toContain(model)
+    expect(getModelPricing(model)?.input).toBeGreaterThan(0)
+  })
+})
+
 describe('catalog featured model metadata', () => {
   it('defines at most one active featured model per provider', () => {
     for (const provider of Object.values(PROVIDER_DEFINITIONS)) {
@@ -292,18 +392,34 @@ describe('sakana provider definition', () => {
     expect(sakana.modelPatterns).toEqual([/^fugu/])
   })
 
-  it('exposes fugu and fugu-ultra with a 1M context window', () => {
-    expect(sakana.models.map((m) => m.id)).toEqual(['fugu', 'fugu-ultra'])
-    for (const model of sakana.models) {
-      expect(model.contextWindow).toBe(1000000)
-    }
+  it('preserves existing aliases and exposes current versioned models', () => {
+    expect(sakana.models.map((model) => model.id)).toEqual(
+      expect.arrayContaining([
+        'fugu',
+        'fugu-ultra',
+        'fugu-ultra-v2.0',
+        'fugu-max',
+        'fugu-max-v1.0',
+        'sakana-namazu',
+        'sakana-namazu-v1.0',
+      ])
+    )
+    expect(sakana.models.find((model) => model.id === 'fugu')?.pricing).toMatchObject({
+      input: 5,
+      output: 30,
+      cachedInput: 0.5,
+      updatedAt: '2026-06-22',
+    })
   })
 
-  it('prices both models at the documented fugu-ultra ceiling', () => {
-    for (const model of sakana.models) {
-      expect(model.pricing.input).toBe(5)
-      expect(model.pricing.output).toBe(30)
-      expect(model.pricing.cachedInput).toBe(0.5)
+  it('keeps current Ultra aliases aligned with the versioned long-context rate', () => {
+    for (const id of ['fugu-ultra', 'fugu-ultra-v2.0']) {
+      expect(sakana.models.find((model) => model.id === id)?.pricing).toMatchObject({
+        input: 5,
+        output: 30,
+        cachedInput: 0.5,
+        tiers: [{ aboveInputTokens: 272000, input: 10, cachedInput: 1, output: 45 }],
+      })
     }
   })
 
@@ -311,6 +427,8 @@ describe('sakana provider definition', () => {
     const baseModels = getBaseModelProviders()
     expect(baseModels.fugu).toBe('sakana')
     expect(baseModels['fugu-ultra']).toBe('sakana')
+    expect(baseModels['fugu-max-v1.0']).toBe('sakana')
+    expect(baseModels['sakana-namazu-v1.0']).toBe('sakana')
   })
 })
 
@@ -318,6 +436,7 @@ describe('nvidia provider definition', () => {
   const nvidia = PROVIDER_DEFINITIONS.nvidia
 
   const expectedModels = [
+    { id: 'nvidia/nemotron-3.5-lightning-30b-a3b', contextWindow: 1000000 },
     { id: 'nvidia/llama-3.1-nemotron-70b-instruct', contextWindow: 128000 },
     { id: 'nvidia/llama-3.1-nemotron-ultra-253b-v1', contextWindow: 131072 },
     { id: 'nvidia/llama-3.3-nemotron-super-49b-v1.5', contextWindow: 131072 },
@@ -333,7 +452,7 @@ describe('nvidia provider definition', () => {
     expect(nvidia.modelPatterns).toEqual([/^nvidia\//])
   })
 
-  it('exposes all six Nemotron models with the documented context windows', () => {
+  it('exposes Nemotron models with the documented context windows', () => {
     expect(nvidia.models.map((m) => m.id)).toEqual(expectedModels.map((m) => m.id))
     for (const expected of expectedModels) {
       const model = nvidia.models.find((m) => m.id === expected.id)
diff --git a/apps/sim/providers/models.ts b/apps/sim/providers/models.ts
index 37ccde6c448..edd140d3b28 100644
--- a/apps/sim/providers/models.ts
+++ b/apps/sim/providers/models.ts
@@ -635,6 +635,23 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         releaseDate: '2026-03-17',
         speedOptimized: true,
       },
+      {
+        id: 'gpt-5.3-codex',
+        pricing: {
+          input: 1.75,
+          cachedInput: 0.175,
+          output: 14.0,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          reasoningEffort: {
+            values: ['low', 'medium', 'high', 'xhigh'],
+          },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 400000,
+        releaseDate: '2026-02-05',
+      },
       // GPT-5.2 family
       {
         id: 'gpt-5.2-pro',
@@ -770,6 +787,20 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         contextWindow: 400000,
         releaseDate: '2025-08-07',
       },
+      {
+        id: 'chat-latest',
+        pricing: {
+          input: 5.0,
+          cachedInput: 0.5,
+          output: 30.0,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 400000,
+        releaseDate: '2026-05-05',
+      },
       {
         id: 'gpt-5-chat-latest',
         pricing: {
@@ -1245,6 +1276,132 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
     icon: AzureIcon,
     isReseller: true,
     models: [
+      {
+        id: 'azure/gpt-6-astra',
+        pricing: {
+          input: 10,
+          cachedInput: 1,
+          output: 50,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          reasoningEffort: {
+            values: ['low', 'medium', 'high', 'xhigh', 'max'],
+          },
+          verbosity: {
+            values: ['low', 'medium', 'high'],
+          },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1050000,
+        releaseDate: '2026-09-03',
+      },
+      {
+        id: 'azure/gpt-5.6-sol',
+        pricing: {
+          input: 4,
+          cachedInput: 0.5,
+          output: 20,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          reasoningEffort: {
+            values: ['none', 'low', 'medium', 'high', 'xhigh', 'max'],
+          },
+          verbosity: {
+            values: ['low', 'medium', 'high'],
+          },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1050000,
+        releaseDate: '2026-07-09',
+      },
+      {
+        id: 'azure/gpt-5.6-terra',
+        pricing: {
+          input: 2,
+          cachedInput: 0.2,
+          output: 12,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          reasoningEffort: {
+            values: ['none', 'low', 'medium', 'high', 'xhigh', 'max'],
+          },
+          verbosity: {
+            values: ['low', 'medium', 'high'],
+          },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1050000,
+        releaseDate: '2026-07-09',
+      },
+      {
+        id: 'azure/gpt-5.6-luna',
+        pricing: {
+          input: 0.2,
+          cachedInput: 0.02,
+          output: 1.2,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          reasoningEffort: {
+            values: ['none', 'low', 'medium', 'high', 'xhigh', 'max'],
+          },
+          verbosity: {
+            values: ['low', 'medium', 'high'],
+          },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1050000,
+        releaseDate: '2026-07-09',
+      },
+      {
+        id: 'azure/gpt-5.5',
+        pricing: {
+          input: 5,
+          cachedInput: 0.5,
+          output: 30,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          reasoningEffort: {
+            values: ['none', 'low', 'medium', 'high', 'xhigh'],
+          },
+          verbosity: {
+            values: ['low', 'medium', 'high'],
+          },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1050000,
+        releaseDate: '2026-04-24',
+      },
+      {
+        id: 'azure/gpt-5.4-pro',
+        pricing: {
+          input: 30,
+          output: 180,
+          tiers: [
+            {
+              aboveInputTokens: 272000,
+              input: 60,
+              output: 270,
+            },
+          ],
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          reasoningEffort: {
+            values: ['medium', 'high', 'xhigh'],
+          },
+          verbosity: {
+            values: ['low', 'medium', 'high'],
+          },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1050000,
+        releaseDate: '2026-03-05',
+      },
       {
         id: 'azure/gpt-4o',
         pricing: {
@@ -1563,6 +1720,90 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
       promptCaching: { minimumCacheableTokens: 1024 },
     },
     models: [
+      {
+        id: 'azure-anthropic/claude-fable-5-1',
+        pricing: {
+          input: 10,
+          cachedInput: 0.25,
+          output: 50,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          forcedToolUse: false,
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 128000,
+          promptCaching: { minimumCacheableTokens: 512 },
+          thinking: {
+            levels: ['low', 'medium', 'high', 'xhigh'],
+            default: 'high',
+            streamed: 'summary',
+          },
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-09-01',
+      },
+      {
+        id: 'azure-anthropic/claude-opus-5',
+        pricing: {
+          input: 5,
+          cachedInput: 0.5,
+          output: 25,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 128000,
+          promptCaching: { minimumCacheableTokens: 512 },
+          thinking: {
+            levels: ['low', 'medium', 'high', 'xhigh', 'max'],
+            default: 'high',
+            streamed: 'summary',
+          },
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-07-24',
+      },
+      {
+        id: 'azure-anthropic/claude-opus-4-8',
+        pricing: {
+          input: 5,
+          cachedInput: 0.5,
+          output: 25,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 128000,
+          thinking: {
+            levels: ['low', 'medium', 'high', 'xhigh', 'max'],
+            default: 'high',
+            streamed: 'summary',
+          },
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-05-28',
+      },
+      {
+        id: 'azure-anthropic/claude-opus-4-7',
+        pricing: {
+          input: 5,
+          cachedInput: 0.5,
+          output: 25,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 128000,
+          promptCaching: { minimumCacheableTokens: 2048 },
+          thinking: {
+            levels: ['low', 'medium', 'high', 'xhigh', 'max'],
+            default: 'high',
+            streamed: 'summary',
+          },
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-04-16',
+      },
       {
         id: 'azure-anthropic/claude-opus-4-6',
         pricing: {
@@ -1607,6 +1848,47 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         contextWindow: 200000,
         releaseDate: '2025-11-24',
       },
+      {
+        id: 'azure-anthropic/claude-sonnet-5',
+        pricing: {
+          input: 2,
+          cachedInput: 0.2,
+          output: 10,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 128000,
+          thinking: {
+            levels: ['low', 'medium', 'high', 'xhigh', 'max'],
+            default: 'high',
+            streamed: 'summary',
+          },
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-06-30',
+      },
+      {
+        id: 'azure-anthropic/claude-sonnet-4-6',
+        pricing: {
+          input: 3,
+          cachedInput: 0.3,
+          output: 15,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 1 },
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 128000,
+          thinking: {
+            levels: ['low', 'medium', 'high', 'max'],
+            default: 'high',
+            streamed: 'summary',
+          },
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-02-17',
+      },
       {
         id: 'azure-anthropic/claude-sonnet-4-5',
         pricing: {
@@ -1707,13 +1989,32 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         featured: true,
         recommended: true,
       },
+      {
+        id: 'gemini-3.7-flash',
+        pricing: {
+          input: 0.75,
+          cachedInput: 0.075,
+          output: 3.75,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 2 },
+          thinking: {
+            levels: ['low', 'medium', 'high'],
+            default: 'medium',
+          },
+          maxOutputTokens: 65536,
+        },
+        contextWindow: 1048576,
+        releaseDate: '2026-08-13',
+      },
       {
         id: 'gemini-3.6-flash',
         pricing: {
-          input: 1.5,
-          cachedInput: 0.15,
-          output: 7.5,
-          updatedAt: '2026-07-21',
+          input: 0.75,
+          cachedInput: 0.075,
+          output: 3.75,
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
@@ -1771,7 +2072,15 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           input: 2.0,
           cachedInput: 0.2,
           output: 12.0,
-          updatedAt: '2026-06-11',
+          tiers: [
+            {
+              aboveInputTokens: 200000,
+              input: 4.0,
+              cachedInput: 0.4,
+              output: 18.0,
+            },
+          ],
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
@@ -1829,7 +2138,15 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           input: 1.25,
           cachedInput: 0.125,
           output: 10.0,
-          updatedAt: '2026-06-11',
+          tiers: [
+            {
+              aboveInputTokens: 200000,
+              input: 2.5,
+              cachedInput: 0.25,
+              output: 15.0,
+            },
+          ],
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
@@ -1942,6 +2259,79 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
       toolUsageControl: true,
     },
     models: [
+      {
+        id: 'vertex/gemini-3.8-flash',
+        pricing: {
+          input: 0.75,
+          cachedInput: 0.075,
+          output: 3.75,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          thinking: {
+            levels: ['low', 'medium', 'high'],
+            default: 'medium',
+          },
+          maxOutputTokens: 65536,
+        },
+        contextWindow: 1048576,
+        releaseDate: '2026-09-02',
+      },
+      {
+        id: 'vertex/gemini-3.7-flash',
+        pricing: {
+          input: 0.75,
+          cachedInput: 0.075,
+          output: 3.75,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          thinking: {
+            levels: ['low', 'medium', 'high'],
+            default: 'medium',
+          },
+          maxOutputTokens: 65536,
+        },
+        contextWindow: 1048576,
+        releaseDate: '2026-08-13',
+      },
+      {
+        id: 'vertex/gemini-3.6-flash',
+        pricing: {
+          input: 0.75,
+          cachedInput: 0.075,
+          output: 3.75,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          thinking: {
+            levels: ['minimal', 'low', 'medium', 'high'],
+            default: 'medium',
+          },
+          maxOutputTokens: 65536,
+        },
+        contextWindow: 1048576,
+        releaseDate: '2026-07-21',
+      },
+      {
+        id: 'vertex/gemini-3.5-flash-lite',
+        pricing: {
+          input: 0.3,
+          cachedInput: 0.03,
+          output: 2.5,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          thinking: {
+            levels: ['minimal', 'low', 'medium', 'high'],
+            default: 'minimal',
+          },
+          maxOutputTokens: 65536,
+        },
+        contextWindow: 1048576,
+        releaseDate: '2026-07-21',
+        speedOptimized: true,
+      },
       {
         id: 'vertex/gemini-3.5-flash',
         pricing: {
@@ -2159,14 +2549,15 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
       {
         id: 'deepseek-v4-pro',
         pricing: {
-          input: 0.435,
-          cachedInput: 0.003625,
-          output: 0.87,
-          updatedAt: '2026-06-16',
+          /** Peak rates; off-peak API requests cost half these rates. */
+          input: 1.32,
+          cachedInput: 0.044,
+          output: 3.96,
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           reasoningEffort: {
-            values: ['high', 'max'],
+            values: ['low', 'high', 'max'],
           },
           thinking: {
             levels: ['none', 'enabled'],
@@ -2177,18 +2568,42 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         contextWindow: 1000000,
         releaseDate: '2026-04-24',
       },
+      {
+        id: 'deepseek-flash',
+        pricing: {
+          /** Peak rates; off-peak API requests cost half these rates. */
+          input: 0.3,
+          cachedInput: 0.006,
+          output: 1.2,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 2 },
+          reasoningEffort: {
+            values: ['low', 'high', 'max'],
+          },
+          thinking: {
+            levels: ['none', 'enabled'],
+            default: 'enabled',
+          },
+          maxOutputTokens: 384000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-09-10',
+      },
       {
         id: 'deepseek-v4-flash',
         pricing: {
-          input: 0.14,
-          cachedInput: 0.0028,
-          output: 0.28,
-          updatedAt: '2026-06-16',
+          /** Legacy alias now serves V4.1 Flash, including its peak pricing. */
+          input: 0.3,
+          cachedInput: 0.006,
+          output: 1.2,
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
           reasoningEffort: {
-            values: ['high', 'max'],
+            values: ['low', 'high', 'max'],
           },
           thinking: {
             levels: ['none', 'enabled'],
@@ -2289,7 +2704,15 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           input: 2.0,
           cachedInput: 0.5,
           output: 6.0,
-          updatedAt: '2026-08-12',
+          tiers: [
+            {
+              aboveInputTokens: 200000,
+              input: 4.0,
+              cachedInput: 1.0,
+              output: 12.0,
+            },
+          ],
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
@@ -2529,6 +2952,23 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
       toolUsageControl: true,
     },
     models: [
+      {
+        id: 'cerebras/qwen-3.8-27b',
+        pricing: {
+          input: 0.99,
+          output: 1.49,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 2 },
+          reasoningEffort: {
+            values: ['none', 'low', 'medium', 'high'],
+          },
+          maxOutputTokens: 40960,
+        },
+        contextWindow: 131072,
+        releaseDate: '2026-08-14',
+      },
       {
         id: 'cerebras/gpt-oss-120b',
         pricing: {
@@ -2681,6 +3121,23 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         releaseDate: '2025-04-29',
         sunset: { status: 'deprecated' },
       },
+      {
+        id: 'groq/qwen/qwen3.8-27b',
+        pricing: {
+          input: 0.8,
+          output: 4.0,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 2 },
+          reasoningEffort: {
+            values: ['none', 'low', 'medium', 'high'],
+          },
+          maxOutputTokens: 16384,
+        },
+        contextWindow: 131042,
+        releaseDate: '2026-08-14',
+      },
       {
         id: 'groq/qwen/qwen3.6-27b',
         pricing: {
@@ -2689,7 +3146,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2026-07-10',
         },
         capabilities: {
-          maxOutputTokens: 32768,
+          maxOutputTokens: 16384,
           thinking: {
             levels: ['enabled'],
             default: 'enabled',
@@ -2756,16 +3213,69 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
   sakana: {
     id: 'sakana',
     name: 'Sakana AI',
-    description: "Sakana AI's Fugu multi-agent models via an OpenAI-compatible API",
+    description: 'Sakana AI Fugu and Namazu models via an OpenAI-compatible API',
     defaultModel: 'fugu',
     modelPatterns: [/^fugu/],
     icon: SakanaIcon,
     color: '#E60000',
     capabilities: {
-      temperature: { min: 0, max: 2 },
       toolUsageControl: true,
     },
     models: [
+      {
+        id: 'fugu-ultra-v2.0',
+        pricing: {
+          input: 5,
+          cachedInput: 0.5,
+          output: 30,
+          tiers: [{ aboveInputTokens: 272000, input: 10, cachedInput: 1, output: 45 }],
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {},
+        releaseDate: '2026-09-11',
+      },
+      {
+        id: 'fugu-max',
+        pricing: {
+          input: 2,
+          cachedInput: 0.25,
+          output: 6,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {},
+        releaseDate: '2026-09-11',
+      },
+      {
+        id: 'fugu-max-v1.0',
+        pricing: {
+          input: 2,
+          cachedInput: 0.25,
+          output: 6,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {},
+        releaseDate: '2026-09-11',
+      },
+      {
+        id: 'sakana-namazu',
+        pricing: {
+          input: 0.95,
+          cachedInput: 0.15,
+          output: 4,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {},
+      },
+      {
+        id: 'sakana-namazu-v1.0',
+        pricing: {
+          input: 0.95,
+          cachedInput: 0.15,
+          output: 4,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {},
+      },
       {
         id: 'fugu',
         pricing: {
@@ -2785,7 +3295,8 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           input: 5,
           cachedInput: 0.5,
           output: 30,
-          updatedAt: '2026-06-22',
+          tiers: [{ aboveInputTokens: 272000, input: 10, cachedInput: 1, output: 45 }],
+          updatedAt: '2026-09-14',
         },
         capabilities: {},
         contextWindow: 1000000,
@@ -2805,10 +3316,22 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
     isReseller: true,
     contextInformationAvailable: true,
     capabilities: {
-      temperature: { min: 0, max: 2 },
       toolUsageControl: true,
     },
     models: [
+      {
+        id: 'nvidia/nemotron-3.5-lightning-30b-a3b',
+        pricing: {
+          input: 0,
+          output: 0,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: false,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-08-11',
+      },
       {
         id: 'nvidia/llama-3.1-nemotron-70b-instruct',
         pricing: {
@@ -2976,6 +3499,9 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           toolUsageControl: true,
+          reasoningEffort: {
+            values: ['low', 'high', 'max'],
+          },
           maxOutputTokens: 1048576,
         },
         contextWindow: 1048576,
@@ -3048,8 +3574,9 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         id: 'glm-5.3',
         pricing: {
           input: 1.4,
+          cachedInput: 0.26,
           output: 4.4,
-          updatedAt: '2026-08-26',
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
@@ -3067,13 +3594,17 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         id: 'glm-5.3-flash',
         pricing: {
           input: 0.15,
+          cachedInput: 0.03,
           output: 0.5,
-          updatedAt: '2026-08-26',
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
           toolUsageControl: true,
           maxOutputTokens: 131072,
+          reasoningEffort: {
+            values: ['low', 'high', 'max'],
+          },
         },
         contextWindow: 1000000,
         releaseDate: '2026-08-26',
@@ -3311,6 +3842,65 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
       toolUsageControl: true,
     },
     models: [
+      {
+        id: 'mistral-medium-3-5',
+        pricing: {
+          input: 1.5,
+          cachedInput: 0.15,
+          output: 7.5,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 1.5 },
+        },
+        contextWindow: 256000,
+        releaseDate: '2026-04-28',
+      },
+      {
+        id: 'mistral-medium-3',
+        pricing: {
+          input: 1.5,
+          cachedInput: 0.15,
+          output: 7.5,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 1.5 },
+        },
+        contextWindow: 256000,
+        releaseDate: '2026-04-28',
+      },
+      {
+        id: 'zai-glm-5-2',
+        pricing: {
+          input: 1.4,
+          cachedInput: 0.14,
+          output: 4.4,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 1.5 },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-08-06',
+      },
+      {
+        /** Free preview retires September 30, 2026: https://docs.mistral.ai/resources/changelogs */
+        id: 'labs-leanstral-1-5',
+        pricing: {
+          input: 0,
+          cachedInput: 0,
+          output: 0,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 1.5 },
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 256000,
+        releaseDate: '2026-06-30',
+      },
       {
         id: 'mistral-large-latest',
         pricing: {
@@ -3435,15 +4025,16 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
       {
         id: 'mistral-medium-latest',
         pricing: {
-          input: 0.4,
-          output: 2.0,
-          updatedAt: '2026-06-11',
+          input: 1.5,
+          cachedInput: 0.15,
+          output: 7.5,
+          updatedAt: '2026-09-14',
         },
         capabilities: {
           temperature: { min: 0, max: 1.5 },
         },
-        contextWindow: 128000,
-        releaseDate: '2025-08-12',
+        contextWindow: 256000,
+        releaseDate: '2026-04-28',
       },
       {
         id: 'mistral-medium-2604',
@@ -3696,7 +4287,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
     name: 'Ollama',
     description: 'Local LLM models via Ollama',
     defaultModel: '',
-    modelPatterns: [],
+    modelPatterns: [/^ollama\//],
     icon: OllamaIcon,
     capabilities: {
       toolUsageControl: false, // Ollama does not support tool_choice parameter
@@ -3714,10 +4305,111 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
     color: '#FF9900',
     isReseller: true,
     capabilities: {
-      temperature: { min: 0, max: 1 },
       toolUsageControl: true,
     },
     models: [
+      {
+        id: 'bedrock/anthropic.claude-opus-5',
+        pricing: {
+          input: 5.5,
+          cachedInput: 0.55,
+          output: 27.5,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-07-24',
+      },
+      {
+        id: 'bedrock/anthropic.claude-sonnet-5',
+        pricing: {
+          input: 2.2,
+          cachedInput: 0.22,
+          output: 11,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-06-30',
+      },
+      {
+        id: 'bedrock/anthropic.claude-fable-5',
+        pricing: {
+          input: 11,
+          cachedInput: 1.1,
+          output: 55,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-06-09',
+      },
+      {
+        id: 'bedrock/anthropic.claude-opus-4-8',
+        pricing: {
+          input: 5.5,
+          cachedInput: 0.55,
+          output: 27.5,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-05-28',
+      },
+      {
+        id: 'bedrock/anthropic.claude-opus-4-7',
+        pricing: {
+          input: 5.5,
+          cachedInput: 0.55,
+          output: 27.5,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-04-16',
+      },
+      {
+        id: 'bedrock/anthropic.claude-sonnet-4-6',
+        pricing: {
+          input: 3.3,
+          cachedInput: 0.33,
+          output: 16.5,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 1 },
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 64000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-02-17',
+      },
+      {
+        id: 'bedrock/anthropic.claude-opus-4-6-v1',
+        pricing: {
+          input: 5.5,
+          cachedInput: 0.55,
+          output: 27.5,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          temperature: { min: 0, max: 1 },
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1000000,
+        releaseDate: '2026-02-05',
+      },
       {
         id: 'bedrock/anthropic.claude-opus-4-5-20251101-v1:0',
         pricing: {
@@ -3785,6 +4477,140 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         releaseDate: '2025-08-05',
         sunset: { status: 'legacy' },
       },
+      {
+        id: 'bedrock/openai.gpt-6-astra',
+        pricing: {
+          input: 11,
+          output: 55,
+          tiers: [{ aboveInputTokens: 272000, input: 22, output: 82.5 }],
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 1050000,
+        releaseDate: '2026-09-08',
+      },
+      {
+        id: 'bedrock/openai.gpt-5.6-sol',
+        pricing: {
+          input: 4.4,
+          output: 22,
+          tiers: [{ aboveInputTokens: 272000, input: 8.8, output: 33 }],
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {},
+        contextWindow: 1000000,
+        releaseDate: '2026-07-13',
+      },
+      {
+        id: 'bedrock/openai.gpt-5.6-terra',
+        pricing: {
+          input: 2.2,
+          output: 13.2,
+          tiers: [{ aboveInputTokens: 272000, input: 4.4, output: 19.8 }],
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {},
+        contextWindow: 1000000,
+        releaseDate: '2026-07-13',
+      },
+      {
+        id: 'bedrock/openai.gpt-5.6-luna',
+        pricing: {
+          input: 0.22,
+          output: 1.32,
+          tiers: [{ aboveInputTokens: 272000, input: 0.44, output: 1.98 }],
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {},
+        contextWindow: 1000000,
+        releaseDate: '2026-07-13',
+      },
+      {
+        id: 'bedrock/openai.gpt-oss-120b-1:0',
+        pricing: {
+          input: 0.15,
+          output: 0.6,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 16000,
+        },
+        contextWindow: 128000,
+        releaseDate: '2025-08-05',
+      },
+      {
+        id: 'bedrock/openai.gpt-oss-20b-1:0',
+        pricing: {
+          input: 0.07,
+          output: 0.3,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 16000,
+        },
+        contextWindow: 128000,
+        releaseDate: '2025-08-05',
+      },
+      {
+        id: 'bedrock/minimax.minimax-m2.5',
+        pricing: {
+          input: 0.3,
+          output: 1.2,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 8000,
+        },
+        contextWindow: 196000,
+        releaseDate: '2026-02-12',
+      },
+      {
+        id: 'bedrock/zai.glm-5',
+        pricing: {
+          input: 1,
+          output: 3.2,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 128000,
+        },
+        contextWindow: 200000,
+        releaseDate: '2026-02-11',
+      },
+      {
+        id: 'bedrock/moonshotai.kimi-k2.5',
+        pricing: {
+          input: 0.6,
+          output: 3,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 16000,
+        },
+        contextWindow: 256000,
+        releaseDate: '2026-01-27',
+      },
+      {
+        id: 'bedrock/deepseek.v3.2',
+        pricing: {
+          input: 0.62,
+          output: 1.85,
+          updatedAt: '2026-09-14',
+        },
+        capabilities: {
+          nativeStructuredOutputs: true,
+          maxOutputTokens: 8000,
+        },
+        contextWindow: 164000,
+        releaseDate: '2025-12-01',
+      },
       {
         id: 'bedrock/amazon.nova-2-lite-v1:0',
         pricing: {
@@ -4233,6 +5059,18 @@ export const DYNAMIC_MODEL_PROVIDERS = [
   'baseten',
 ] as const
 
+/** Providers whose adapter accepts namespaced IDs, rather than only a native model-family name. */
+const NAMESPACED_MODEL_PROVIDERS = new Set(
+  Object.entries(PROVIDER_DEFINITIONS)
+    .filter(
+      ([providerId, provider]) =>
+        provider.isReseller ||
+        (DYNAMIC_MODEL_PROVIDERS as readonly string[]).includes(providerId) ||
+        provider.models.some((model) => model.id.startsWith(`${providerId}/`))
+    )
+    .map(([providerId]) => providerId)
+)
+
 function getAllStaticModelIds(): string[] {
   const ids: string[] = []
   for (const [providerId, provider] of Object.entries(PROVIDER_DEFINITIONS)) {
@@ -4253,7 +5091,7 @@ export function isKnownModelId(modelId: string): boolean {
 
   const lowered = trimmed.toLowerCase()
   for (const provider of DYNAMIC_MODEL_PROVIDERS) {
-    if (lowered.startsWith(`${provider}/`)) return true
+    if (lowered.startsWith(`${provider}/`) && lowered.slice(provider.length + 1).trim()) return true
   }
 
   return false
@@ -4347,9 +5185,20 @@ export function getBaseModelProviders(): Record<string, ProviderId> {
     )
 }
 
-export function getProviderFromModel(model: string): ProviderId {
+/** Resolves catalog entries and provider patterns without guessing a fallback provider. */
+export function findProviderFromModel(model: string): ProviderId | null {
   const normalizedModel = model.toLowerCase()
 
+  /** Explicit provider namespaces take precedence over names discovered on a local server. */
+  for (const [providerId, provider] of Object.entries(PROVIDER_DEFINITIONS)) {
+    if (
+      NAMESPACED_MODEL_PROVIDERS.has(providerId) &&
+      provider.modelPatterns?.some((pattern) => pattern.test(normalizedModel))
+    ) {
+      return providerId as ProviderId
+    }
+  }
+
   for (const [providerId, provider] of Object.entries(PROVIDER_DEFINITIONS)) {
     if (
       provider.models.some((providerModel) => providerModel.id.toLowerCase() === normalizedModel)
@@ -4364,7 +5213,20 @@ export function getProviderFromModel(model: string): ProviderId {
     }
   }
 
-  return 'ollama'
+  return null
+}
+
+export function getProviderFromModel(model: string): ProviderId {
+  return findProviderFromModel(model) ?? 'ollama'
+}
+
+/** Recognized provider namespaces accept deployment and model IDs outside the static catalog. */
+export function isCustomModelId(modelId: string): boolean {
+  const separator = modelId.indexOf('/')
+  if (separator < 1 || modelId.slice(separator + 1).trim().length === 0) return false
+
+  const providerId = findProviderFromModel(modelId)
+  return providerId !== null && NAMESPACED_MODEL_PROVIDERS.has(providerId)
 }
 
 export function getProviderIcon(model: string): React.ComponentType<{ className?: string }> | null {
diff --git a/apps/sim/providers/nvidia/index.ts b/apps/sim/providers/nvidia/index.ts
index 5973ba2af54..4f29d730527 100644
--- a/apps/sim/providers/nvidia/index.ts
+++ b/apps/sim/providers/nvidia/index.ts
@@ -6,7 +6,11 @@ import type { ChatCompletionChunk } from 'openai/resources/chat/completions'
 import type { StreamingExecution } from '@/executor/types'
 import { MAX_TOOL_ITERATIONS } from '@/providers'
 import { formatMessagesForProvider } from '@/providers/attachments'
-import { getProviderDefaultModel, getProviderModels } from '@/providers/models'
+import {
+  getModelCapabilities,
+  getProviderDefaultModel,
+  getProviderModels,
+} from '@/providers/models'
 import { createReadableStreamFromNvidiaStream } from '@/providers/nvidia/utils'
 import { createOpenAICompatAssistantHistory } from '@/providers/openai-compat/assistant-history'
 import { executeProviderTool } from '@/providers/runtime-context'
@@ -25,6 +29,7 @@ import type {
 import { ProviderError } from '@/providers/types'
 import {
   calculateCost,
+  generateSchemaInstructions,
   isFunctionToolCall,
   prepareToolExecution,
   prepareToolsWithUsageControl,
@@ -86,13 +91,22 @@ export const nvidiaProvider: ProviderConfig = {
         allMessages.push(...request.messages)
       }
       const formattedMessages = formatMessagesForProvider(allMessages, 'nvidia')
+      const useJsonMode =
+        !!request.responseFormat &&
+        getModelCapabilities(request.model)?.nativeStructuredOutputs === false
+      if (useJsonMode) {
+        formattedMessages.push({
+          role: 'system',
+          content: generateSchemaInstructions(request.responseFormat),
+        })
+      }
 
       const tools = request.tools?.length
         ? request.tools.map((tool) => adaptOpenAIChatToolSchema(tool))
         : undefined
 
       const payload: any = {
-        model: request.model,
+        model: request.model.replace(/^nvidia\//i, 'nvidia/'),
         messages: formattedMessages,
       }
 
@@ -100,16 +114,20 @@ export const nvidiaProvider: ProviderConfig = {
       if (request.maxTokens != null) payload.max_tokens = request.maxTokens
 
       const responseFormatPayload = request.responseFormat
-        ? {
-            type: 'json_schema' as const,
-            json_schema: {
-              name: request.responseFormat.name || 'response_schema',
-              schema: request.responseFormat.schema || request.responseFormat,
-              strict: request.responseFormat.strict !== false,
-            },
-          }
+        ? useJsonMode
+          ? { type: 'json_object' as const }
+          : {
+              type: 'json_schema' as const,
+              json_schema: {
+                name: request.responseFormat.name || 'response_schema',
+                schema: request.responseFormat.schema || request.responseFormat,
+                strict: request.responseFormat.strict !== false,
+              },
+            }
         : undefined
 
+      if (useJsonMode) payload.chat_template_kwargs = { enable_thinking: false }
+
       let preparedTools: ReturnType<typeof prepareToolsWithUsageControl> | null = null
       let hasActiveTools = false
 
diff --git a/apps/sim/providers/ollama-cloud/index.test.ts b/apps/sim/providers/ollama-cloud/index.test.ts
index cbd33346641..ceed999a97a 100644
--- a/apps/sim/providers/ollama-cloud/index.test.ts
+++ b/apps/sim/providers/ollama-cloud/index.test.ts
@@ -184,6 +184,18 @@ describe('ollamaCloudProvider.executeRequest', () => {
     expect(result).toMatchObject({ content: 'hello', model: 'gpt-oss:120b' })
   })
 
+  it.each([
+    ['ollama-cloud/deepseek-v4.1-flash', 'deepseek-v4.1-flash'],
+    ['ollama-cloud/glm-5.3', 'glm-5.3'],
+    ['OLLAMA-CLOUD/Org/CustomModel', 'Org/CustomModel'],
+  ])(
+    'forwards new and custom cloud models without changing their IDs: %s',
+    async (model, wireModel) => {
+      await ollamaCloudProvider.executeRequest({ ...baseRequest, model })
+      expect(mockCreate.mock.calls[0][0].model).toBe(wireModel)
+    }
+  )
+
   it('assembles system, context, then history in order and forwards params', async () => {
     await ollamaCloudProvider.executeRequest({
       ...baseRequest,
diff --git a/apps/sim/providers/ollama-cloud/index.ts b/apps/sim/providers/ollama-cloud/index.ts
index 11d7bb4372b..b23f7ccc8f2 100644
--- a/apps/sim/providers/ollama-cloud/index.ts
+++ b/apps/sim/providers/ollama-cloud/index.ts
@@ -28,7 +28,7 @@ export const ollamaCloudProvider: ProviderConfig = {
       throw new Error('API key is required for Ollama Cloud')
     }
 
-    const requestedModel = request.model.replace(/^ollama-cloud\//, '')
+    const requestedModel = request.model.replace(/^ollama-cloud\//i, '')
 
     return executeOllamaProviderRequest(
       { ...request, model: requestedModel },
diff --git a/apps/sim/providers/ollama/index.test.ts b/apps/sim/providers/ollama/index.test.ts
index feda81c9afa..d3940ca3fe7 100644
--- a/apps/sim/providers/ollama/index.test.ts
+++ b/apps/sim/providers/ollama/index.test.ts
@@ -149,6 +149,14 @@ describe('ollamaProvider.executeRequest', () => {
     mockExecuteTool.mockResolvedValue({ success: true, output: { ok: true } })
   })
 
+  it.each(['ollama/Org/CustomModel', 'OLLAMA/Org/CustomModel', 'Org/CustomModel'])(
+    'preserves the local model ID while removing only its optional namespace: %s',
+    async (model) => {
+      await ollamaProvider.executeRequest({ ...baseRequest, model })
+      expect(mockCreate.mock.calls[0][0].model).toBe('Org/CustomModel')
+    }
+  )
+
   it('assembles system, context, then history in order and forwards params', async () => {
     const result = (await ollamaProvider.executeRequest({
       ...baseRequest,
diff --git a/apps/sim/providers/ollama/index.ts b/apps/sim/providers/ollama/index.ts
index cd0e16f8bd7..fac7a49a8d1 100644
--- a/apps/sim/providers/ollama/index.ts
+++ b/apps/sim/providers/ollama/index.ts
@@ -48,17 +48,20 @@ export const ollamaProvider: ProviderConfig = {
   executeRequest: async (
     request: ProviderRequest
   ): Promise<ProviderResponse | StreamingExecution> => {
-    return executeOllamaProviderRequest(request, {
-      providerId: 'ollama',
-      providerLabel: 'Ollama',
-      createClient: () =>
-        new OpenAI({
-          ...openAICompatTransport(),
-          apiKey: 'empty',
-          baseURL: `${OLLAMA_HOST}/v1`,
-        }),
-      createStream: createReadableStreamFromOllamaStream,
-      logger,
-    })
+    return executeOllamaProviderRequest(
+      { ...request, model: request.model.replace(/^ollama\//i, '') },
+      {
+        providerId: 'ollama',
+        providerLabel: 'Ollama',
+        createClient: () =>
+          new OpenAI({
+            ...openAICompatTransport(),
+            apiKey: 'empty',
+            baseURL: `${OLLAMA_HOST}/v1`,
+          }),
+        createStream: createReadableStreamFromOllamaStream,
+        logger,
+      }
+    )
   },
 }
diff --git a/apps/sim/providers/openrouter/index.test.ts b/apps/sim/providers/openrouter/index.test.ts
index 4af5747f4a3..694606065fb 100644
--- a/apps/sim/providers/openrouter/index.test.ts
+++ b/apps/sim/providers/openrouter/index.test.ts
@@ -174,6 +174,12 @@ describe('openRouterProvider.executeRequest', () => {
     expect(payload.messages.at(-1)).toEqual({ role: 'user', content: 'Hello' })
   })
 
+  it('preserves custom provider paths when stripping an uppercase namespace', async () => {
+    mockCreate.mockResolvedValueOnce(textResponse('ok'))
+    await openRouterProvider.executeRequest({ ...baseRequest, model: 'OPENROUTER/Org/CustomModel' })
+    expect(mockCreate.mock.calls[0][0].model).toBe('Org/CustomModel')
+  })
+
   it('inserts context as a user message between system and history', async () => {
     mockCreate.mockResolvedValueOnce(textResponse('ok'))
 
diff --git a/apps/sim/providers/openrouter/index.ts b/apps/sim/providers/openrouter/index.ts
index fbfeddd6193..2ca900a9905 100644
--- a/apps/sim/providers/openrouter/index.ts
+++ b/apps/sim/providers/openrouter/index.ts
@@ -107,7 +107,7 @@ export const openRouterProvider: ProviderConfig = {
       baseURL: 'https://openrouter.ai/api/v1',
     })
 
-    const requestedModel = request.model.replace(/^openrouter\//, '')
+    const requestedModel = request.model.replace(/^openrouter\//i, '')
 
     logger.info('Preparing OpenRouter request', {
       model: requestedModel,
diff --git a/apps/sim/providers/openrouter/utils.ts b/apps/sim/providers/openrouter/utils.ts
index 2f8a7850fc9..c6763c6b514 100644
--- a/apps/sim/providers/openrouter/utils.ts
+++ b/apps/sim/providers/openrouter/utils.ts
@@ -77,7 +77,7 @@ export async function getOpenRouterModelCapabilities(
     cacheTimestamp = now
   }
 
-  const normalizedId = modelId.replace(/^openrouter\//, '')
+  const normalizedId = modelId.replace(/^openrouter\//i, '')
   return modelCapabilitiesCache.get(normalizedId) ?? null
 }
 
diff --git a/apps/sim/providers/sakana/index.ts b/apps/sim/providers/sakana/index.ts
index 33c9861a626..4648bdc7a72 100644
--- a/apps/sim/providers/sakana/index.ts
+++ b/apps/sim/providers/sakana/index.ts
@@ -7,6 +7,7 @@ import type { StreamingExecution } from '@/executor/types'
 import { MAX_TOOL_ITERATIONS } from '@/providers'
 import { formatMessagesForProvider } from '@/providers/attachments'
 import { getProviderDefaultModel, getProviderModels } from '@/providers/models'
+import { createOpenAICompatAssistantHistory } from '@/providers/openai-compat/assistant-history'
 import { executeProviderTool } from '@/providers/runtime-context'
 import { createReadableStreamFromSakanaStream } from '@/providers/sakana/utils'
 import { createSettledAgentEventStream } from '@/providers/stream-events'
@@ -38,7 +39,7 @@ const SAKANA_BASE_URL = 'https://api.sakana.ai/v1'
 export const sakanaProvider: ProviderConfig = {
   id: 'sakana',
   name: 'Sakana AI',
-  description: "Sakana AI's Fugu multi-agent models via an OpenAI-compatible API",
+  description: 'Sakana AI Fugu and Namazu models via an OpenAI-compatible API',
   version: '1.0.0',
   models: getProviderModels('sakana'),
   defaultModel: getProviderDefaultModel('sakana'),
@@ -339,18 +340,16 @@ export const sakanaProvider: ProviderConfig = {
 
           const executionResults = await Promise.all(toolExecutionPromises)
 
-          currentMessages.push({
-            role: 'assistant',
-            content: null,
-            tool_calls: toolCallsInResponse.map((tc) => ({
-              id: tc.id,
-              type: 'function',
-              function: {
-                name: tc.function.name,
-                arguments: tc.function.arguments,
-              },
-            })),
-          })
+          const assistantMessage = currentResponse.choices[0]?.message
+          if (assistantMessage) {
+            currentMessages.push(
+              createOpenAICompatAssistantHistory({
+                message: assistantMessage,
+                toolCalls: toolCallsInResponse,
+                reasoningFields: ['reasoning_content'],
+              })
+            )
+          }
 
           for (const executionResult of executionResults) {
             const { toolCall, toolName, toolParams, result, startTime, endTime, duration } =
diff --git a/apps/sim/providers/settled-tool-streams.test.ts b/apps/sim/providers/settled-tool-streams.test.ts
index 44f4d4622ab..69fbfdae77a 100644
--- a/apps/sim/providers/settled-tool-streams.test.ts
+++ b/apps/sim/providers/settled-tool-streams.test.ts
@@ -48,7 +48,11 @@ vi.mock('@/providers/models', () => ({
     .fn()
     .mockReturnValue({ maxBytes: 10 * 1024 * 1024, strategy: 'inline' }),
   INLINE_ATTACHMENT_MAX_BYTES: 10 * 1024 * 1024,
-  getModelCapabilities: vi.fn(),
+  getModelCapabilities: vi.fn((model: string) =>
+    model === 'nvidia/nemotron-3.5-lightning-30b-a3b'
+      ? { nativeStructuredOutputs: false }
+      : undefined
+  ),
   getProviderModels: vi.fn((provider: string) => [`${provider}/test-model`]),
   getProviderDefaultModel: vi.fn((provider: string) => `${provider}/test-model`),
 }))
@@ -181,6 +185,12 @@ const PROVIDERS = [
 ] as const
 
 const REASONING_HISTORY_PROVIDERS = [
+  {
+    name: 'Sakana Namazu',
+    provider: sakanaProvider,
+    model: 'sakana-namazu-v1.0',
+    field: 'reasoning_content',
+  },
   {
     name: 'Cerebras',
     provider: cerebrasProvider,
@@ -238,6 +248,13 @@ const CAPPED_PROVIDERS = [
 ] as const
 
 const STRUCTURED_OUTPUT_PROVIDERS = [
+  {
+    name: 'NVIDIA Lightning',
+    provider: nvidiaProvider,
+    model: 'nvidia/nemotron-3.5-lightning-30b-a3b',
+    responseFormatType: 'json_object',
+    disablesTools: 'none',
+  },
   {
     name: 'Baseten',
     provider: basetenProvider,
@@ -389,6 +406,36 @@ describe('settled provider tool streams', () => {
     mockExecuteTool.mockResolvedValue({ success: true, output: { value: 'found' } })
   })
 
+  it('normalizes the NVIDIA namespace while preserving the upstream model name', async () => {
+    mockCreate.mockResolvedValueOnce(response('ok'))
+    await nvidiaProvider.executeRequest({
+      apiKey: 'test-key',
+      model: 'NVIDIA/CustomModel',
+      messages: [{ role: 'user', content: 'Hello' }],
+    })
+    expect(mockCreate.mock.calls[0][0].model).toBe('nvidia/CustomModel')
+  })
+
+  it('uses NVIDIA Lightning JSON mode with schema instructions and reasoning disabled', async () => {
+    mockCreate.mockResolvedValueOnce(response('{"value":"found"}'))
+
+    await nvidiaProvider.executeRequest({
+      apiKey: 'test-key',
+      model: 'nvidia/nemotron-3.5-lightning-30b-a3b',
+      messages: [{ role: 'user', content: 'Return a value' }],
+      responseFormat: {
+        name: 'result',
+        schema: { type: 'object', properties: { value: { type: 'string' } } },
+      },
+    })
+
+    expect(mockCreate.mock.calls[0][0]).toMatchObject({
+      response_format: { type: 'json_object' },
+      chat_template_kwargs: { enable_thinking: false },
+      messages: expect.arrayContaining([{ role: 'system', content: 'SCHEMA_INSTRUCTIONS' }]),
+    })
+  })
+
   it.each(PROVIDERS)(
     '$name projects the existing final answer without another provider call',
     async ({ provider, model }) => {
diff --git a/apps/sim/providers/specialist-reasoning.test.ts b/apps/sim/providers/specialist-reasoning.test.ts
new file mode 100644
index 00000000000..5122cd6398f
--- /dev/null
+++ b/apps/sim/providers/specialist-reasoning.test.ts
@@ -0,0 +1,164 @@
+/**
+ * @vitest-environment node
+ */
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+import type { ProviderRequest } from '@/providers/types'
+
+const { mockCreate } = vi.hoisted(() => ({ mockCreate: vi.fn() }))
+
+vi.mock('openai', () => ({
+  default: vi.fn().mockImplementation(
+    class {
+      chat = { completions: { create: mockCreate } }
+    }
+  ),
+}))
+
+vi.mock('@cerebras/cerebras_cloud_sdk', () => ({
+  Cerebras: vi.fn().mockImplementation(
+    class {
+      chat = { completions: { create: mockCreate } }
+    }
+  ),
+}))
+
+vi.mock('@/providers', () => ({ MAX_TOOL_ITERATIONS: 3 }))
+vi.mock('@/providers/attachments', () => ({
+  formatMessagesForProvider: vi.fn((messages) => messages),
+}))
+vi.mock('@/providers/trace-enrichment', () => ({
+  enrichLastModelSegmentFromChatCompletions: vi.fn(),
+}))
+vi.mock('@/providers/runtime-context', () => ({
+  executeProviderTool: vi.fn().mockResolvedValue({
+    rawResponse: { success: true, output: { result: 'found' } },
+    modelResponse: { success: true, output: { result: 'found' } },
+  }),
+}))
+
+import { cerebrasProvider } from '@/providers/cerebras'
+import { kimiProvider } from '@/providers/kimi'
+
+function request(model: string, overrides: Partial<ProviderRequest> = {}): ProviderRequest {
+  return {
+    model,
+    apiKey: 'test-key',
+    messages: [{ role: 'user', content: 'hello' }],
+    ...overrides,
+  }
+}
+
+describe('specialist provider reasoning parameters', () => {
+  beforeEach(() => {
+    mockCreate.mockReset()
+    mockCreate.mockResolvedValue({
+      choices: [{ message: { content: 'ok', tool_calls: [] } }],
+      usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
+    })
+  })
+
+  it.each(['none', 'low', 'medium', 'high'] as const)(
+    'Cerebras forwards Qwen 3.8 effort %s',
+    async (reasoningEffort) => {
+      await cerebrasProvider.executeRequest(
+        request('cerebras/qwen-3.8-27b', { reasoningEffort, maxTokens: 40960 })
+      )
+      expect(mockCreate.mock.calls[0][0]).toMatchObject({
+        model: 'qwen-3.8-27b',
+        reasoning_effort: reasoningEffort,
+        max_completion_tokens: 40960,
+      })
+    }
+  )
+
+  it.each(['low', 'high', 'max'] as const)(
+    'Kimi K3 forwards effort %s',
+    async (reasoningEffort) => {
+      await kimiProvider.executeRequest(request('kimi-k3', { reasoningEffort, temperature: 0.2 }))
+      const payload = mockCreate.mock.calls[0][0]
+      expect(payload.reasoning_effort).toBe(reasoningEffort)
+      expect(payload.temperature).toBeUndefined()
+      expect(payload.thinking).toBeUndefined()
+    }
+  )
+
+  it.each([
+    { provider: cerebrasProvider, model: 'cerebras/qwen-3.8-27b' },
+    { provider: kimiProvider, model: 'kimi-k3' },
+  ])('preserves the server default for $model', async ({ provider, model }) => {
+    await provider.executeRequest(request(model, { reasoningEffort: 'auto' }))
+    expect(mockCreate.mock.calls[0][0].reasoning_effort).toBeUndefined()
+  })
+
+  it('preserves custom Cerebras identifiers while removing only the leading prefix', async () => {
+    await cerebrasProvider.executeRequest(request('Cerebras/Organization/Model-A'))
+    expect(mockCreate.mock.calls[0][0].model).toBe('Organization/Model-A')
+  })
+
+  it('K3 requires each forced tool, then restores all tools with automatic choice', async () => {
+    for (const name of ['search', 'lookup']) {
+      mockCreate.mockResolvedValueOnce({
+        choices: [
+          {
+            message: {
+              content: null,
+              reasoning_content: `Use ${name}`,
+              tool_calls: [
+                { id: `call-${name}`, type: 'function', function: { name, arguments: '{}' } },
+              ],
+            },
+          },
+        ],
+        usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
+      })
+    }
+    await kimiProvider.executeRequest(
+      request('kimi-k3', {
+        tools: ['search', 'lookup', 'optional'].map((id) => ({
+          id,
+          name: id,
+          description: id,
+          params: {},
+          parameters: { type: 'object', properties: {} },
+          usageControl: id === 'optional' ? 'auto' : 'force',
+        })),
+      })
+    )
+    const payloads = mockCreate.mock.calls.map(([payload]) => payload)
+    expect(payloads).toHaveLength(3)
+    expect(payloads[0].tool_choice).toBe('required')
+    expect(
+      payloads[0].tools.map((tool: { function: { name: string } }) => tool.function.name)
+    ).toEqual(['search'])
+    expect(payloads[1].tool_choice).toBe('required')
+    expect(
+      payloads[1].tools.map((tool: { function: { name: string } }) => tool.function.name)
+    ).toEqual(['lookup'])
+    expect(payloads[2].tool_choice).toBe('auto')
+    expect(payloads[2].tools).toHaveLength(3)
+    expect(payloads[1].messages).toContainEqual(
+      expect.objectContaining({ reasoning_content: 'Use search' })
+    )
+  })
+
+  it.each(['kimi-k2.7-code', 'kimi-k2.7-code-highspeed'])(
+    '%s keeps automatic tool choice when forced selection is unsupported',
+    async (model) => {
+      await kimiProvider.executeRequest(
+        request(model, {
+          tools: [
+            {
+              id: 'lookup',
+              description: 'Lookup',
+              params: {},
+              parameters: {},
+              usageControl: 'force',
+            },
+          ],
+        })
+      )
+      expect(mockCreate.mock.calls[0][0].tool_choice).toBe('auto')
+      expect(mockCreate.mock.calls[0][0].thinking).toBeUndefined()
+    }
+  )
+})
diff --git a/apps/sim/providers/together/index.test.ts b/apps/sim/providers/together/index.test.ts
index c9c75846cb9..123da1e9c03 100644
--- a/apps/sim/providers/together/index.test.ts
+++ b/apps/sim/providers/together/index.test.ts
@@ -154,6 +154,14 @@ describe('togetherProvider', () => {
     expect(result).toHaveProperty('execution')
   })
 
+  it('preserves custom model casing after an uppercase provider prefix', async () => {
+    mockCreate.mockResolvedValueOnce(textResponse('ok'))
+
+    await togetherProvider.executeRequest({ ...baseRequest, model: 'TOGETHER/Org/Custom-Model' })
+
+    expect(callBody(0).model).toBe('Org/Custom-Model')
+  })
+
   it('sends a json_schema response_format with no strict field', async () => {
     mockCreate.mockResolvedValueOnce(textResponse('{}'))
 
diff --git a/apps/sim/providers/together/index.ts b/apps/sim/providers/together/index.ts
index 27d0f7ed769..55047796976 100644
--- a/apps/sim/providers/together/index.ts
+++ b/apps/sim/providers/together/index.ts
@@ -92,7 +92,7 @@ export const togetherProvider: ProviderConfig = {
       baseURL: 'https://api.together.ai/v1',
     })
 
-    const requestedModel = request.model.replace(/^together\//, '')
+    const requestedModel = request.model.replace(/^together\//i, '')
 
     logger.info('Preparing Together request', {
       model: requestedModel,
diff --git a/apps/sim/providers/utils.test.ts b/apps/sim/providers/utils.test.ts
index 14fbe4d68d1..a3c148277e0 100644
--- a/apps/sim/providers/utils.test.ts
+++ b/apps/sim/providers/utils.test.ts
@@ -37,6 +37,7 @@ import {
   getReasoningEffortValuesForModel,
   getThinkingLevelsForModel,
   getVerbosityValuesForModel,
+  isGemini3Model,
   isProviderBlacklisted,
   MODELS_TEMP_RANGE_0_1,
   MODELS_TEMP_RANGE_0_2,
@@ -57,6 +58,7 @@ import {
   transformBlockTool,
   updateOllamaProviderModels,
 } from '@/providers/utils'
+import { useProvidersStore } from '@/stores/providers/store'
 
 const mockGetRotatingApiKey = vi.fn().mockReturnValue('rotating-server-key')
 const originalRequire = module.require
@@ -163,6 +165,25 @@ describe('getApiKey', () => {
     expect(key2).toBe('empty')
   })
 
+  it.each(['ollama', 'vllm', 'litellm'] as const)(
+    'uses the routed cloud provider credentials despite a name collision in %s discovery',
+    (localProvider) => {
+      const originalProviders = useProvidersStore.getState().providers
+      useProvidersStore.setState({
+        providers: {
+          ...originalProviders,
+          [localProvider]: { ...originalProviders[localProvider], models: ['azure/MyDeployment'] },
+        },
+      })
+      try {
+        expect(getApiKey('azure-openai', 'azure/MyDeployment', 'azure-key')).toBe('azure-key')
+        expect(() => getApiKey('azure-openai', 'azure/MyDeployment')).toThrow('API key is required')
+      } finally {
+        useProvidersStore.setState({ providers: originalProviders })
+      }
+    }
+  )
+
   it('should return empty or user-provided key for vllm provider without requiring API key', () => {
     setEnvFlags({ isHosted: false })
 
@@ -546,10 +567,11 @@ describe('Model Capabilities', () => {
         (m) =>
           m.includes('gpt-5') &&
           !m.includes('chat-latest') &&
-          !m.includes('gpt-5.5-pro') &&
-          !m.includes('gpt-5.4-pro') &&
-          !m.includes('gpt-5.2-pro') &&
-          !m.includes('gpt-5-pro')
+          m !== 'gpt-5.5-pro' &&
+          m !== 'gpt-5.4-pro' &&
+          m !== 'gpt-5.3-codex' &&
+          m !== 'gpt-5.2-pro' &&
+          m !== 'gpt-5-pro'
       )
       const gpt5ModelsWithVerbosity = MODELS_WITH_VERBOSITY.filter(
         (m) => m.includes('gpt-5') && !m.includes('chat-latest')
@@ -562,6 +584,9 @@ describe('Model Capabilities', () => {
       expect(MODELS_WITH_REASONING_EFFORT).toContain('gpt-5.4-pro')
       expect(MODELS_WITH_VERBOSITY).not.toContain('gpt-5.4-pro')
 
+      expect(MODELS_WITH_REASONING_EFFORT).toContain('gpt-5.3-codex')
+      expect(MODELS_WITH_VERBOSITY).not.toContain('gpt-5.3-codex')
+
       expect(MODELS_WITH_REASONING_EFFORT).toContain('gpt-5.2-pro')
       expect(MODELS_WITH_VERBOSITY).not.toContain('gpt-5.2-pro')
 
@@ -828,6 +853,27 @@ describe('Cost Calculation', () => {
       expect(longContext).toMatchObject({ input: 1.088004, output: 1.8, total: 2.888004 })
     })
 
+    it.each([
+      ['gemini-3.1-pro-preview', 2, 0.2, 12, 4, 0.4, 18],
+      ['gemini-2.5-pro', 1.25, 0.125, 10, 2.5, 0.25, 15],
+      ['grok-4.6', 2, 0.5, 6, 4, 1, 12],
+    ])(
+      'applies %s long-context rates only above 200k prompt tokens, including cached input',
+      (model, input, cached, output, longInput, longCached, longOutput) => {
+        const shortContext = calculateCost(model, 200_000, 100_000)
+        const longContext = calculateCost(model, 200_001, 100_000)
+        const shortCached = calculateCost(model, 200_000, 100_000, true)
+        const longCachedCost = calculateCost(model, 200_001, 100_000, true)
+
+        expect(shortContext.input).toBeCloseTo(input * 0.2, 10)
+        expect(shortContext.output).toBeCloseTo(output * 0.1, 10)
+        expect(longContext.input).toBeCloseTo((longInput * 200_001) / 1e6, 10)
+        expect(longContext.output).toBeCloseTo(longOutput * 0.1, 10)
+        expect(shortCached.input).toBeCloseTo(cached * 0.2, 10)
+        expect(longCachedCost.input).toBeCloseTo((longCached * 200_001) / 1e6, 10)
+      }
+    )
+
     it('should return default pricing for unknown models', () => {
       const result = calculateCost('unknown-model', 1000, 500, false)
 
@@ -2105,6 +2151,18 @@ describe('describeModelLevel', () => {
 })
 
 describe('findProviderFromModel', () => {
+  it.each([
+    ['azure/MyDeployment', 'azure-openai'],
+    ['AZURE/MyDeployment', 'azure-openai'],
+    ['azure-anthropic/MyDeployment', 'azure-anthropic'],
+    ['bedrock/custom-inference-profile', 'bedrock'],
+    ['vertex/publishers/google/models/custom-gemini', 'vertex'],
+  ])('uses the declared provider namespace for %s', (model, provider) => {
+    expect(findProviderFromModel(model)).toBe(provider)
+    expect(getProviderFromModel(model)).toBe(provider)
+    expect(shouldBillModelUsage(model)).toBe(false)
+  })
+
   it('resolves a chat model to its declaring provider', () => {
     expect(findProviderFromModel('claude-sonnet-5')).toBe('anthropic')
     expect(findProviderFromModel('gpt-5.2')).toBe('openai')
@@ -2128,6 +2186,26 @@ describe('findProviderFromModel', () => {
   })
 })
 
+describe('isGemini3Model', () => {
+  it.each([
+    'gemini-3.8-flash',
+    'VERTEX/gemini-3.8-flash',
+    'vertex/google/gemini-3.8-flash',
+    'vertex/publishers/google/models/gemini-3.8-flash',
+    'vertex/projects/test-project/locations/global/publishers/google/models/gemini-3.8-flash',
+  ])('recognizes the Gemini family in %s', (model) => {
+    expect(isGemini3Model(model)).toBe(true)
+  })
+
+  it.each([
+    'vertex/gemini-2.5-pro',
+    'vertex/custom-gemini-3-deployment',
+    'vertex/publishers/another-provider/models/gemini-3.8-flash',
+  ])('does not infer Gemini 3 behavior from %s', (model) => {
+    expect(isGemini3Model(model)).toBe(false)
+  })
+})
+
 describe('transformBlockTool param decoding', () => {
   /**
    * `StoredTool.params` stringifies every value, so a tool row hands a block the same
diff --git a/apps/sim/providers/utils.ts b/apps/sim/providers/utils.ts
index ba9d19a2728..be2d64b46dd 100644
--- a/apps/sim/providers/utils.ts
+++ b/apps/sim/providers/utils.ts
@@ -25,6 +25,7 @@ import { assembleCustomBlockInputMapping, isCustomBlockType } from '@/blocks/cus
 import type { SubBlockConfig } from '@/blocks/types'
 import { isCustomTool } from '@/executor/constants'
 import {
+  findProviderFromModel as findProviderFromDefinitions,
   getComputerUseModels,
   getHostedModels as getHostedModelsFromDefinitions,
   getMaxOutputTokensForModel as getMaxOutputTokensForModelFromDefinitions,
@@ -60,7 +61,6 @@ import {
   registerPreparedProviderToolInputProvenance,
 } from '@/providers/tool-input-provenance'
 import type { ModelPricing, ProviderId, ProviderToolConfig } from '@/providers/types'
-import { useProvidersStore } from '@/stores/providers/store'
 import { mergeToolParameters } from '@/tools/merge-params'
 import { buildToolParamShapes, decodeToolParams } from '@/tools/param-shape'
 import type { WorkflowToolExecutionContext } from '@/tools/types'
@@ -282,18 +282,7 @@ export function getAllModelProviders(): Record<string, ProviderId> {
  * that was never about it.
  */
 export function findProviderFromModel(model: string): ProviderId | null {
-  const normalizedModel = model.toLowerCase()
-
-  const declared = getAllModelProviders()[normalizedModel]
-  if (declared) return declared
-
-  for (const [id, config] of Object.entries(providers)) {
-    for (const pattern of config.modelPatterns ?? []) {
-      if (pattern.test(normalizedModel)) return id as ProviderId
-    }
-  }
-
-  return null
+  return findProviderFromDefinitions(model)
 }
 
 export function getProviderFromModel(model: string): ProviderId {
@@ -1163,27 +1152,20 @@ export const PROVIDER_PLACEHOLDER_KEY = 'provider-uses-own-credentials'
 export function getApiKey(provider: string, model: string, userProvidedKey?: string): string {
   const hasUserKey = !!userProvidedKey
 
-  const isOllamaModel =
-    provider === 'ollama' || useProvidersStore.getState().providers.ollama.models.includes(model)
-  if (isOllamaModel) {
+  if (provider === 'ollama') {
     return 'empty'
   }
 
-  const isVllmModel =
-    provider === 'vllm' || useProvidersStore.getState().providers.vllm.models.includes(model)
-  if (isVllmModel) {
+  if (provider === 'vllm') {
     return userProvidedKey || 'empty'
   }
 
-  const isLitellmModel =
-    provider === 'litellm' || useProvidersStore.getState().providers.litellm.models.includes(model)
-  if (isLitellmModel) {
+  if (provider === 'litellm') {
     return userProvidedKey || 'empty'
   }
 
-  // Bedrock uses its own credentials (bedrockAccessKeyId/bedrockSecretKey), not apiKey
-  const isBedrockModel = provider === 'bedrock' || model.startsWith('bedrock/')
-  if (isBedrockModel) {
+  /** Bedrock authenticates through its configured AWS credentials. */
+  if (provider === 'bedrock') {
     return PROVIDER_PLACEHOLDER_KEY
   }
 
@@ -1569,7 +1551,13 @@ export function isDeepResearchModel(model: string): boolean {
 }
 
 export function isGemini3Model(model: string): boolean {
-  const normalized = model.toLowerCase().replace(/^vertex\//, '')
+  const normalized = model
+    .toLowerCase()
+    .replace(/^vertex\//, '')
+    .replace(
+      /^(?:google\/|(?:projects\/[^/]+\/locations\/[^/]+\/)?publishers\/google\/models\/)/,
+      ''
+    )
   return normalized.startsWith('gemini-3')
 }
 
diff --git a/apps/sim/providers/vertex/index.test.ts b/apps/sim/providers/vertex/index.test.ts
index 35282292617..424dcee8c5b 100644
--- a/apps/sim/providers/vertex/index.test.ts
+++ b/apps/sim/providers/vertex/index.test.ts
@@ -25,10 +25,6 @@ vi.mock('google-auth-library', () => ({
   },
 }))
 vi.mock('@/providers/gemini/core', () => ({ executeGeminiRequest: mockExecuteGeminiRequest }))
-vi.mock('@/providers/models', () => ({
-  getProviderModels: () => ['vertex/gemini-2.0-flash'],
-  getProviderDefaultModel: () => 'vertex/gemini-2.0-flash',
-}))
 vi.mock('@/lib/core/config/env', () => ({ env: {} }))
 
 import { vertexProvider } from '@/providers/vertex'
@@ -102,4 +98,43 @@ describe('vertexProvider location and project validation', () => {
 
     expect(genAIArgs[0]).toMatchObject({ location: 'us-central1' })
   })
+
+  it.each([
+    'vertex/gemini-3.8-flash',
+    'vertex/gemini-3.7-flash',
+    'vertex/gemini-3.6-flash',
+    'vertex/gemini-3.5-flash-lite',
+    'vertex/publishers/google/models/gemini-3.8-flash',
+  ])('defaults %s to the supported global endpoint', async (model) => {
+    await vertexProvider.executeRequest(request({ model }))
+
+    expect(genAIArgs[0]).toMatchObject({ location: 'global' })
+  })
+
+  it.each(['us', 'eu', 'global'])(
+    'preserves the explicitly selected %s endpoint for Gemini 3',
+    async (vertexLocation) => {
+      await vertexProvider.executeRequest(
+        request({ model: 'vertex/gemini-3.8-flash', vertexLocation })
+      )
+
+      expect(genAIArgs[0]).toMatchObject({ location: vertexLocation })
+    }
+  )
+
+  it.each([
+    ['VERTEX/Custom-Deployment', 'Custom-Deployment'],
+    ['vertex/publishers/google/models/Custom-Model', 'publishers/google/models/Custom-Model'],
+    [
+      'vertex/projects/MyProject/locations/global/publishers/google/models/Custom-Model',
+      'projects/MyProject/locations/global/publishers/google/models/Custom-Model',
+    ],
+    ['publishers/vertex/models/Custom-Model', 'publishers/vertex/models/Custom-Model'],
+  ])('preserves the custom model identifier in %s', async (model, expectedModel) => {
+    await vertexProvider.executeRequest(request({ model }))
+
+    expect(mockExecuteGeminiRequest).toHaveBeenCalledWith(
+      expect.objectContaining({ model: expectedModel, providerType: 'vertex' })
+    )
+  })
 })
diff --git a/apps/sim/providers/vertex/index.ts b/apps/sim/providers/vertex/index.ts
index 59c22efc2ad..65e49151d81 100644
--- a/apps/sim/providers/vertex/index.ts
+++ b/apps/sim/providers/vertex/index.ts
@@ -10,6 +10,7 @@ import type { StreamingExecution } from '@/executor/types'
 import { executeGeminiRequest } from '@/providers/gemini/core'
 import { getProviderDefaultModel, getProviderModels } from '@/providers/models'
 import type { ProviderConfig, ProviderRequest, ProviderResponse } from '@/providers/types'
+import { isGemini3Model } from '@/providers/utils'
 
 const logger = createLogger('VertexProvider')
 
@@ -34,13 +35,13 @@ export const vertexProvider: ProviderConfig = {
   executeRequest: async (
     request: ProviderRequest
   ): Promise<ProviderResponse | StreamingExecution> => {
+    const model = request.model.replace(/^vertex\//i, '')
     const vertexProject = request.vertexProject || env.VERTEX_PROJECT
-    // Hostnames are case-insensitive, so a mixed-case location reaches Google fine
-    // today. Normalize before validating rather than rejecting it as malformed.
+    /** Gemini 3 models use global or multi-region endpoints instead of legacy regions. */
     const vertexLocation = (
       request.vertexLocation ||
       env.VERTEX_LOCATION ||
-      'us-central1'
+      (isGemini3Model(model) ? 'global' : 'us-central1')
     ).toLowerCase()
 
     if (!vertexProject) {
@@ -71,9 +72,6 @@ export const vertexProvider: ProviderConfig = {
       )
     }
 
-    // Strip 'vertex/' prefix from model name if present
-    const model = request.model.replace('vertex/', '')
-
     logger.info('Creating Vertex AI client', {
       project: vertexProject,
       location: vertexLocation,
diff --git a/apps/sim/providers/vllm/index.test.ts b/apps/sim/providers/vllm/index.test.ts
index 1057bb89e5e..29459978c6d 100644
--- a/apps/sim/providers/vllm/index.test.ts
+++ b/apps/sim/providers/vllm/index.test.ts
@@ -150,6 +150,15 @@ describe('vllmProvider', () => {
     mockCreatePinnedFetch.mockReturnValue(pinnedFetchFn)
   })
 
+  it('preserves a custom served-model name when stripping an uppercase namespace', async () => {
+    mockCreate.mockResolvedValueOnce(chatResponse('hello'))
+    await vllmProvider.executeRequest({
+      model: 'VLLM/Org/CustomModel',
+      messages: [{ role: 'user', content: 'hi' }],
+    })
+    expect(createPayload(0).model).toBe('Org/CustomModel')
+  })
+
   describe('endpoint SSRF protection', () => {
     it('does not validate or pin when no endpoint is supplied (uses env base URL)', async () => {
       mockCreate.mockResolvedValueOnce(chatResponse('hi'))
diff --git a/apps/sim/providers/vllm/index.ts b/apps/sim/providers/vllm/index.ts
index 3bad1cba441..d1394a94230 100644
--- a/apps/sim/providers/vllm/index.ts
+++ b/apps/sim/providers/vllm/index.ts
@@ -181,7 +181,7 @@ export const vllmProvider: ProviderConfig = {
       : undefined
 
     const payload: any = {
-      model: request.model.replace(/^vllm\//, ''),
+      model: request.model.replace(/^vllm\//i, ''),
       messages: formattedMessages,
     }