BofAI · Will-Guan · Jul 7, 2026 · Jul 6, 2026 · Jul 7, 2026 · Jul 7, 2026
diff --git a/docs/llmservice/models/glm-5-1.md b/docs/llmservice/models/glm-5-1.md
@@ -2,7 +2,7 @@
 
 ## Overview
 
-GLM-5.1 is an open-source flagship AI model developed by Z.ai, formerly Zhipu AI, a Tsinghua University spinoff and the first publicly traded foundation model company. Released on April 7, 2026, it is a post-training upgrade to GLM-5, built on a 754-billion-parameter Mixture-of-Experts architecture with 40 billion active parameters per token. GLM-5.1 is designed for agentic engineering and long-horizon autonomous software development.
+GLM-5.1 is an open-source flagship AI model developed by Z.ai, formerly Zhipu AI, a Tsinghua University spinoff and the first publicly traded foundation model company. Released on April 7, 2026, it is a post-training upgrade in the GLM family, built on a 754-billion-parameter Mixture-of-Experts architecture with 40 billion active parameters per token. GLM-5.1 is designed for agentic engineering and long-horizon autonomous software development.
 
 ## Key Features
 
@@ -23,7 +23,7 @@ GLM-5.1 is an open-source flagship AI model developed by Z.ai, formerly Zhipu AI
 | :----------------- | :----------------------------------------------------------------------------------------------------------- |
 | **Reasoning**      | AIME 2026: 95.3%, GPQA-Diamond: 86.2%, with strong system-level reasoning across planning and iterative debugging |
 | **Coding**         | SWE-Bench Pro 58.4%, CyberGym 68.7%, BrowseComp 68.0%, MCP-Atlas 71.8%                                      |
-| **Multimodal**     | Text only. No image, audio, or video input. A separate GLM-5V-Turbo variant is available for vision tasks   |
+| **Multimodal**     | Text only. No image, audio, or video input. Vision tasks require a separate vision-capable model             |
 | **Response Speed** | Not independently benchmarked yet; expected to be comparable to similar-scale MoE models                    |
 | **Context Window** | 200K tokens                                                                                                  |
 | **Max Output**     | 128K tokens                                                                                                  |
@@ -32,7 +32,7 @@ GLM-5.1 is an open-source flagship AI model developed by Z.ai, formerly Zhipu AI
 
 ### Known Limitations
 
-* Text-only input with no native multimodal support. Vision use cases rely on the separate GLM-5V-Turbo model.
+* Text-only input with no native multimodal support. Vision use cases rely on a separate vision-capable model.
 * Math and science benchmark scores trail some top proprietary models, making it less suitable for purely quantitative research tasks.
 * On broader coding composites such as Terminal-Bench 2.0 plus NL2Repo, Claude Opus 4.6 still leads.
 * Self-hosting requires substantial compute resources because of the 754B parameter count.

diff --git a/docs/llmservice/models/glm-5-2.md b/docs/llmservice/models/glm-5-2.md
@@ -24,7 +24,7 @@ GLM-5.2 is a GLM-family text foundation model developed by Z.AI and released on
 | **Reasoning**        | Supports deep-thinking mode and `reasoning_effort`; Z.AI positions it for complex engineering, debugging, and long-chain reasoning workflows. |
 | **Creative Writing** | Supports general text generation through the chat completion API, but official GLM-5.2 materials emphasize coding and engineering use cases.  |
 | **Coding**           | Z.AI reports Terminal-Bench 2.1 score of 81.0 and SWE-bench Pro score of 62.1, with focus on long-horizon coding-agent scenarios.             |
-| **Multimodal**       | Text input and text output. Vision and multimodal workflows are handled by separate Z.AI models such as GLM-5V-Turbo.                         |
+| **Multimodal**       | Text input and text output. Vision and multimodal workflows are handled by separate Z.AI vision-language models.                              |
 | **Response Speed**   | Official docs do not publish latency or tokens-per-second figures; streaming responses and streaming tool calls are supported.                |
 | **Context Window**   | 1M tokens.                                                                                                                                    |
 | **Max Output**       | 128K tokens.                                                                                                                                  |
@@ -33,7 +33,7 @@ GLM-5.2 is a GLM-family text foundation model developed by Z.AI and released on
 
 ### Known Limitations
 
-* Text-only model; image, video, and GUI-understanding tasks require a separate vision-language model such as GLM-5V-Turbo.
+* Text-only model; image, video, and GUI-understanding tasks require a separate vision-language model.
 * Very long contexts and 128K outputs can increase latency and cost; cap `max_tokens` and use context caching where applicable.
 
 ## Credits Usage

diff --git a/docs/llmservice/models/glm-5.md b/docs/llmservice/models/glm-5.md
diff --git a/docs/llmservice/models/kimi-k2.6.md b/docs/llmservice/models/kimi-k2.6.md
@@ -0,0 +1,50 @@
+# Kimi K2.6
+
+## Overview
+
+Kimi K2.6 is a Moonshot AI model available on B.AI for multimodal understanding, coding assistance, long-context analysis, and agentic workflows. It is suitable for tasks that combine text, visual context, structured reasoning, and tool-assisted execution. Specific capabilities, input formats, context limits, and availability may vary by B.AI model catalog and platform configuration.
+
+## Key Features
+
+* **Multimodal Understanding**: Supports text and visual-context tasks where enabled by the platform configuration.
+* **Coding and Technical Workflows**: Suitable for coding assistance, debugging, code review, implementation planning, and UI-oriented generation tasks.
+* **Agentic Workflows**: Designed for multi-step tasks that rely on planning, tool use, and iterative execution.
+* **Long-Context Analysis**: Can be used for document review, repository understanding, and multi-turn workflows when long-context support is enabled.
+* **Cost-Aware Usage**: Provides a lower-cost option for multimodal and agentic tasks compared with many higher-priced flagship models.
+
+## Best Use Cases
+
+* **Coding and UI Generation**: Turning requirements, screenshots, and technical context into implementation plans or code suggestions.
+* **Multimodal Document Work**: Understanding documents, diagrams, screenshots, and visual references where supported.
+* **Agent and Tool Workflows**: Multi-step workflows that require reasoning, tool calls, and iterative refinement.
+* **Cost-Sensitive Production Tasks**: High-frequency workloads where price, latency, and model capability need to be balanced.
+
+## Capabilities and Limitations
+
+| Capability         | Description                                                                                      |
+| :----------------- | :----------------------------------------------------------------------------------------------- |
+| **Reasoning**      | Suitable for structured reasoning, technical analysis, and multi-step task planning               |
+| **Coding**         | Supports code generation, debugging, code review, and UI-oriented implementation assistance       |
+| **Multimodal**     | Text and visual input support may be available depending on platform configuration                |
+| **Response Speed** | Optimized for practical interactive and agent workflows; actual latency may vary by workload      |
+| **Context Window** | Long-context support is available where enabled by the platform configuration                     |
+| **Max Output**     | Output limits may vary by platform configuration and request settings                             |
+| **Tool Use**       | Suitable for compatible agent and tool-assisted workflows                                         |
+| **Multilingual**   | Supports multilingual text tasks, including Chinese and English workflows                         |
+
+### Known Limitations
+
+* Specific multimodal input formats, context limits, and output limits may vary by B.AI platform configuration.
+* External actions, web access, code execution, and tool calls require compatible integrations.
+* Model availability and exact runtime behavior may change with the B.AI model catalog and provider-side updates.
+* For the latest supported model ID and runtime availability, refer to the platform model selector or API model catalog.
+
+## Credits Usage
+
+| Model | Input (Credits/Token) | Cache Write (Credits/Token) | Cache Read (Credits/Token) | Output (Credits/Token) | Web Search (Credits/Use) | Billing Notes |
+| :--- | --------------------: | --------------------------: | -------------------------: | ---------------------: | -----------------------: | :--- |
+| **Kimi K2.6** | `0.95` | `0.95` | `0.16` | `4.00` | `-` | - |
+
+:::info Pricing note
+Prices shown in the documentation are B.AI standard reference prices for base billing purposes. B.AI may provide lower actual usage costs through top-up bonuses and account benefits. Specific prices, bonus Credits, and account benefits are subject to the platform display and final billing records.
+:::
diff --git a/docs/llmservice/models/minimax-m2.7.md b/docs/llmservice/models/minimax-m2.7.md
@@ -6,7 +6,7 @@ MiniMax M2.7 is a reasoning-focused large language model developed by MiniMax (S
 
 ## Key Features
 
-* **Cost-Efficient Reasoning**: Achieves intelligence scores comparable to GLM-5 and Kimi K2.5 while costing roughly one-third as much to run, with 20% fewer output tokens needed for equivalent results.
+* **Cost-Efficient Reasoning**: Delivers strong reasoning performance at a lower operating cost than many flagship models, with 20% fewer output tokens needed for equivalent results.
 * **Low Hallucination Rate**: Scores a 34% hallucination rate on the AA-Omniscience Index, lower than Claude Sonnet 4.6 and Gemini 3.1 Pro Preview.
 * **Multi-Agent Collaboration**: Native support for multi-agent orchestration and complex skill coordination, including dynamic tool discovery and invocation at runtime.
 * **Self-Evolution**: Can autonomously complete 30-50% of reinforcement-learning research workflows, representing an early step toward model self-improvement.

diff --git a/docs/llmservice/pricing-and-usage.md b/docs/llmservice/pricing-and-usage.md
@@ -16,11 +16,11 @@ The platform uses a unified Credits system to measure and settle usage across al
 | :---------------- | --------------------: | --------------------------: | -------------------------: | ---------------------: | -----------------------: |
 | MiniMax M3        |                  0.30 |                        0.30 |                       0.06 |                   1.20 |                        - |
 | MiniMax M2.7      |                  0.30 |                       0.375 |                       0.06 |                   1.20 |                        - |
+| Kimi K2.6         |                  0.95 |                        0.95 |                       0.16 |                   4.00 |                        - |
 | Kimi K2.5         |                  0.59 |                        0.59 |                      0.177 |                   3.00 |                        - |
 | Qwen3.6-27B       |                  0.19 |                        0.19 |                      0.019 |                   2.99 |                        - |
 | GLM-5.2           |                  1.40 |                        1.40 |                       0.28 |                   4.40 |                        - |
 | GLM-5.1           |                  1.40 |                        1.40 |                       0.26 |                   4.40 |                        - |
-| GLM-5             |                  1.00 |                        1.00 |                       0.20 |                   3.20 |                        - |
 | DeepSeek V3.2     |                  0.29 |                        0.29 |                      0.145 |                   0.44 |                        - |
 | DeepSeek V4 Flash |                  0.28 |                        0.28 |                     0.0056 |                   0.56 |                        - |
 | DeepSeek V4 Pro   |                  0.87 |                        0.87 |                     0.0087 |                   1.74 |                        - |

diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/glm-5-1.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/glm-5-1.md
@@ -2,7 +2,7 @@
 
 ## 概述
 
-GLM-5.1 是由 Z.ai 开发的开源旗舰 AI 模型。Z.ai 前身为智谱 AI，源自清华大学，也是首家公开上市的基础模型公司。GLM-5.1 于 2026 年 4 月 7 日发布，是在 GLM-5 基础上的一次后训练升级，采用 7540 亿参数的 Mixture-of-Experts 架构，每个 token 激活约 400 亿参数，重点面向 Agent 工程与长周期自主软件开发场景。
+GLM-5.1 是由 Z.ai 开发的开源旗舰 AI 模型。Z.ai 前身为智谱 AI，源自清华大学，也是首家公开上市的基础模型公司。GLM-5.1 于 2026 年 4 月 7 日发布，是 GLM 系列的一次后训练升级，采用 7540 亿参数的 Mixture-of-Experts 架构，每个 token 激活约 400 亿参数，重点面向 Agent 工程与长周期自主软件开发场景。
 
 ## 核心特性
 
@@ -23,7 +23,7 @@ GLM-5.1 是由 Z.ai 开发的开源旗舰 AI 模型。Z.ai 前身为智谱 AI，
 | :--- | :--- |
 | **推理能力** | AIME 2026：95.3%，GPQA-Diamond：86.2%，在规划与迭代调试场景中具备较强的系统级推理能力 |
 | **编程能力** | SWE-Bench Pro 58.4%，CyberGym 68.7%，BrowseComp 68.0%，MCP-Atlas 71.8% |
-| **多模态能力** | 仅支持文本，不支持图像、音频或视频输入；视觉场景可使用单独的 GLM-5V-Turbo 变体 |
+| **多模态能力** | 仅支持文本，不支持图像、音频或视频输入；视觉场景需要使用单独的视觉能力模型 |
 | **响应速度** | 暂无独立公开测速结果，预计与同规模 MoE 模型相近 |
 | **上下文窗口** | 200K tokens |
 | **最大输出** | 128K tokens |
@@ -32,7 +32,7 @@ GLM-5.1 是由 Z.ai 开发的开源旗舰 AI 模型。Z.ai 前身为智谱 AI，
 
 ### 已知限制
 
-* 仅支持文本输入，不具备原生多模态能力；视觉任务需依赖独立的 GLM-5V-Turbo 模型。
+* 仅支持文本输入，不具备原生多模态能力；视觉任务需依赖独立的视觉能力模型。
 * 数学与科学基准成绩仍落后于部分顶级专有模型，因此在纯量化研究任务上不一定是最优选择。
 * 在更广泛的编码综合评测（如 Terminal-Bench 2.0 + NL2Repo）中，Claude Opus 4.6 仍然领先。
 * 由于参数规模达到 754B，自托管需要较高的计算资源。

diff --git a/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/glm-5-2.md b/i18n/zh-Hans/docusaurus-plugin-content-docs/current/llmservice/models/glm-5-2.md
@@ -24,7 +24,7 @@ GLM-5.2 是由 Z.AI 开发的 GLM 系列文本基础模型，于 2026 年 6 月
 | **推理能力** | 支持 deep-thinking 模式和 `reasoning_effort`；Z.AI 将其定位于复杂工程、调试和长链路推理工作流 |
 | **创意写作** | 支持通过 chat completion API 进行通用文本生成，但官方 GLM-5.2 材料更强调代码和工程场景 |
 | **编程能力** | Z.AI 报告 Terminal-Bench 2.1 得分为 81.0，SWE-bench Pro 得分为 62.1，重点面向长周期 Coding Agent 场景 |
-| **多模态能力** | 文本输入和文本输出；视觉和多模态工作流由 GLM-5V-Turbo 等独立 Z.AI 模型处理 |
+| **多模态能力** | 文本输入和文本输出；视觉和多模态工作流由独立的 Z.AI 视觉语言模型处理 |
 | **响应速度** | 官方文档未公布延迟或 tokens-per-second 数据；支持流式响应和流式工具调用 |
 | **上下文窗口** | 1M tokens |
 | **最大输出** | 128K tokens |
@@ -33,7 +33,7 @@ GLM-5.2 是由 Z.AI 开发的 GLM 系列文本基础模型，于 2026 年 6 月
 
 ### 已知限制
 
-* 该模型为文本模型；图像、视频和 GUI 理解任务需要使用 GLM-5V-Turbo 等独立视觉语言模型。
+* 该模型为文本模型；图像、视频和 GUI 理解任务需要使用独立视觉语言模型。
 * 超长上下文和 128K 输出可能增加延迟和成本；建议按需限制 `max_tokens`，并在适用场景使用上下文缓存。
 
 ## 积分消耗