From 91c407267e726e133aed7d0afbc039358b7a621a Mon Sep 17 00:00:00 2001 From: Gezim Hoxha Date: Wed, 2 Sep 2026 01:46:10 +0400 Subject: [PATCH] Add missing Wafer serverless models Adds the five models listed at https://docs.wafer.ai/serverless/setup#models that were missing from providers/wafer.ai: GLM-5.3, GLM-5.3-Flash, Kimi-K3, DeepSeek-V4-Flash-0731-Fast, and Qwen3.5-397B-A17B. Pricing, context, and reasoning-effort tiers taken from the live catalog (GET https://pass.wafer.ai/v1/models), which the docs cite as authoritative. Qwen3.5-397B-A17B has no published Wafer pricing yet, so its cost block is intentionally omitted with a header comment explaining why. Co-Authored-By: Claude Fable 5 --- .../models/DeepSeek-V4-Flash-0731-Fast.toml | 20 ++++++++++++++++++ providers/wafer.ai/models/GLM-5.3-Flash.toml | 21 +++++++++++++++++++ providers/wafer.ai/models/GLM-5.3.toml | 20 ++++++++++++++++++ providers/wafer.ai/models/Kimi-K3.toml | 17 +++++++++++++++ .../wafer.ai/models/Qwen3.5-397B-A17B.toml | 9 ++++++++ 5 files changed, 87 insertions(+) create mode 100644 providers/wafer.ai/models/DeepSeek-V4-Flash-0731-Fast.toml create mode 100644 providers/wafer.ai/models/GLM-5.3-Flash.toml create mode 100644 providers/wafer.ai/models/GLM-5.3.toml create mode 100644 providers/wafer.ai/models/Kimi-K3.toml create mode 100644 providers/wafer.ai/models/Qwen3.5-397B-A17B.toml diff --git a/providers/wafer.ai/models/DeepSeek-V4-Flash-0731-Fast.toml b/providers/wafer.ai/models/DeepSeek-V4-Flash-0731-Fast.toml new file mode 100644 index 00000000000..0e924bd43d4 --- /dev/null +++ b/providers/wafer.ai/models/DeepSeek-V4-Flash-0731-Fast.toml @@ -0,0 +1,20 @@ +# Sources (accessed 2026-09-02): +# - Pricing/capabilities: GET https://pass.wafer.ai/v1/models +# - Model list: https://docs.wafer.ai/serverless/setup#models +# Off is reasoning_effort=none; graded levels none|low|high|max (default low) +base_model = "deepseek/deepseek-v4-flash-0731" +name = "DeepSeek V4 Flash 0731 Fast" +description = "The same model served for high TPS." +reasoning_options = [{ type = "effort", values = ["none", "low", "high", "max"] }] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.20 +output = 0.25 +cache_read = 0.05 +cache_write = 0 + +[limit] +context = 1_048_576 diff --git a/providers/wafer.ai/models/GLM-5.3-Flash.toml b/providers/wafer.ai/models/GLM-5.3-Flash.toml new file mode 100644 index 00000000000..4d4fb3e321e --- /dev/null +++ b/providers/wafer.ai/models/GLM-5.3-Flash.toml @@ -0,0 +1,21 @@ +# Sources (accessed 2026-09-02): +# - Pricing/capabilities: GET https://pass.wafer.ai/v1/models +# - Model list: https://docs.wafer.ai/serverless/setup#models +# Host control differs from the lab API: Wafer self-hosts GLM-5.3-Flash and +# applies its own effort control (system-prompt conditioning); its per-model +# metadata advertises reasoning_effort none|low|high|max (default low, none = +# off; medium is not a distinct tier). +base_model = "zhipuai/glm-5.3-flash" +reasoning_options = [{ type = "effort", values = ["none", "low", "high", "max"] }] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.15 +output = 0.50 +cache_read = 0.03 +cache_write = 0 + +[limit] +context = 1_048_576 diff --git a/providers/wafer.ai/models/GLM-5.3.toml b/providers/wafer.ai/models/GLM-5.3.toml new file mode 100644 index 00000000000..6ff45a31fd3 --- /dev/null +++ b/providers/wafer.ai/models/GLM-5.3.toml @@ -0,0 +1,20 @@ +# Sources (accessed 2026-09-02): +# - Pricing/capabilities: GET https://pass.wafer.ai/v1/models +# - Model list: https://docs.wafer.ai/serverless/setup#models +# Host control differs from the lab API: Wafer self-hosts GLM-5.3 and applies +# its own effort control (system-prompt conditioning); its per-model metadata +# advertises reasoning_effort none|low|high|max (default low), with none = off. +base_model = "zhipuai/glm-5.3" +reasoning_options = [{ type = "effort", values = ["none", "low", "high", "max"] }] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.40 +output = 4.40 +cache_read = 0.26 +cache_write = 0 + +[limit] +context = 1_048_576 diff --git a/providers/wafer.ai/models/Kimi-K3.toml b/providers/wafer.ai/models/Kimi-K3.toml new file mode 100644 index 00000000000..d8089802fe0 --- /dev/null +++ b/providers/wafer.ai/models/Kimi-K3.toml @@ -0,0 +1,17 @@ +# Sources (accessed 2026-09-02): +# - Pricing/capabilities: GET https://pass.wafer.ai/v1/models +# - Model list: https://docs.wafer.ai/serverless/setup#models +# Wafer per-model metadata: "Kimi-K3 reasoning is exposed as off/high/max; +# low and medium collapse to the same enabled tier" — so reasoning_effort is +# none|high|max (default none), and low is deliberately not listed. +base_model = "moonshotai/kimi-k3" +reasoning_options = [{ type = "effort", values = ["none", "high", "max"] }] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 3.00 +output = 15.00 +cache_read = 0.30 +cache_write = 0 diff --git a/providers/wafer.ai/models/Qwen3.5-397B-A17B.toml b/providers/wafer.ai/models/Qwen3.5-397B-A17B.toml new file mode 100644 index 00000000000..eefed5f59c5 --- /dev/null +++ b/providers/wafer.ai/models/Qwen3.5-397B-A17B.toml @@ -0,0 +1,9 @@ +# Sources (accessed 2026-09-02): +# - Model list: https://docs.wafer.ai/serverless/setup#models (262K context, +# serverless-only, non-ZDR) +# - GET https://pass.wafer.ai/v1/models lists the model but publishes no +# pricing or capability metadata for it yet; no public Wafer price as of +# 2026-09-02, so cost is intentionally omitted. +# Toggle: enable_thinking = true|false (normalized by the Wafer API) +base_model = "alibaba/qwen3.5-397b-a17b" +reasoning_options = [{ type = "toggle" }]