chore(sync): update Hugging Face model catalog

[github-actions]

Updates model TOMLs for the huggingface sync target.

Provider	Status	Created	Updated	Deleted
Hugging Face	changed	1	0	0

Hugging Face changed files

• created: providers/huggingface/models/openai/gpt-oss-20b.toml

Notices

Hugging Face

• 73 Hugging Face Inference Providers models were not created because their IDs could not be mapped to provider-agnostic metadata, had no live provider, or had no priced provider.
• Skipped remote IDs: meta-llama/Llama-3.1-8B-Instruct, meta-llama/Meta-Llama-3-8B-Instruct, nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4, Qwen/Qwen3-8B, Qwen/Qwen2.5-7B-Instruct, google/gemma-3-4b-it, nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16, Qwen/Qwen3-4B-Instruct-2507, zai-org/GLM-5.2-FP8, Qwen/Qwen2.5-Coder-32B-Instruct, microsoft/phi-4, Qwen/Qwen2.5-Coder-7B-Instruct, deepseek-ai/DeepSeek-R1-Distill-Qwen-7B, Qwen/Qwen3-235B-A22B-Instruct-2507, Qwen/Qwen3-VL-235B-A22B-Instruct, allenai/Olmo-3-7B-Instruct, google/gemma-3-27b-it, google/gemma-3-12b-it, deepseek-ai/DeepSeek-R1-Distill-Llama-8B, zai-org/GLM-4.6V-Flash, Qwen/Qwen2.5-VL-72B-Instruct, CohereLabs/c4ai-command-r7b-12-2024, swiss-ai/Apertus-8B-Instruct-2509, openai/gpt-oss-safeguard-20b, Qwen/Qwen3-VL-235B-A22B-Thinking, meta-llama/Llama-Guard-4-12B, alpindale/WizardLM-2-8x22B, Qwen/Qwen3-VL-30B-A3B-Instruct, pearl-ai/Gemma-4-31B-it-pearl, Sao10K/L3-8B-Stheno-v3.2, Sao10K/L3-8B-Lunaris-v1, meta-llama/Llama-4-Scout-17B-16E-Instruct, swiss-ai/Apertus-70B-Instruct-2509, speakleash/Bielik-11B-v3.0-Instruct, inclusionAI/Ling-2.6-1T, deepseek-ai/DeepSeek-R1-Distill-Llama-70B, CohereLabs/aya-expanse-32b, CohereLabs/tiny-aya-global, deepseek-ai/DeepSeek-V3, CohereLabs/c4ai-command-r-08-2024, Qwen/Qwen2.5-72B-Instruct, meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8, deepseek-ai/DeepSeek-V3-0324, CohereLabs/c4ai-command-r7b-arabic-02-2025, Qwen/Qwen3-14B, baidu/ERNIE-4.5-VL-424B-A47B-Base-PT, zai-org/GLM-4-32B-0414, CohereLabs/c4ai-command-a-03-2025, MiniMaxAI/MiniMax-M1-80k, CohereLabs/command-a-reasoning-08-2025, deepseek-ai/DeepSeek-V3.1-Terminus, zai-org/AutoGLM-Phone-9B-Multilingual, deepseek-ai/DeepSeek-V3.1, CohereLabs/command-a-translate-08-2025, zai-org/GLM-4.5V-FP8, CohereLabs/tiny-aya-water, meta-llama/Llama-3.1-70B-Instruct, zai-org/GLM-4.6V, CohereLabs/tiny-aya-earth, google/gemma-3n-E4B-it, zai-org/GLM-4.6V-FP8, deepseek-ai/DeepSeek-V3.2-Exp, CohereLabs/aya-vision-32b, deepcogito/cogito-671b-v2.1, aisingapore/Gemma-SEA-LION-v4-27B-IT, CohereLabs/command-a-vision-07-2025, deepcogito/cogito-671b-v2.1-FP8, aisingapore/Qwen-SEA-LION-v4-32B-IT, zai-org/GLM-4.6-FP8, zai-org/GLM-5.1-FP8, Qwen/Qwen2.5-Coder-3B-Instruct, Qwen/Qwen3-4B-Thinking-2507, utter-project/EuroLLM-22B-Instruct-2512
• 6 local Hugging Face models were absent from the Inference Providers catalog and were retained for manual lifecycle review.
• Retained local paths: deepseek-ai/DeepSeek-V4-Pro.toml, XiaomiMiMo/MiMo-V2-Flash.toml, Qwen/Qwen3-Embedding-8B.toml, Qwen/Qwen3-Embedding-4B.toml, Qwen/Qwen3-Next-80B-A3B-Thinking.toml, moonshotai/Kimi-K2-Thinking.toml

This PR was created automatically by the daily model sync workflow.

Evidence

• Hugging Face GPT-OSS guide states that both gpt-oss-20b and gpt-oss-120b are served through Inference Providers and that Chat Completions exposes reasoning effort controls.
• Hugging Face GPT-OSS 20B model card identifies the exact model’s supported levels as low, medium, and high.
• Groq reasoning documentation explicitly documents low, medium, and high for the exact openai/gpt-oss-20b model. Groq is the current fastest live backend in Hugging Face’s /v1/models response and therefore the default route.
• Together OpenAI compatibility documentation independently documents reasoning_effort values low, medium, and high for GPT-OSS models; Together is also a live Hugging Face backend for openai/gpt-oss-20b.

No authenticated request matrix was run because HF_TOKEN was unavailable locally.

Validation

• bun validate
• git diff --check