openai/gpt-oss

Health v2 · maintenance only

Adoption not part of the score

20428 stars · 2160 forks observed · 2026-09-24

What it is AI-extracted, prompt v1, taxonomy v1, 2026-10-19, confidence 0.97

gpt-oss provides open-weight large language models (120B and 20B) with full chain‑of‑thought reasoning and native agentic capabilities. The repository includes reference PyTorch, Triton, and Metal implementations and tools for inference, fine‑tuning, and integration.

Use cases

• Generate high‑quality text and code with a single‑GPU LLM.
• Run reasoning‑heavy tasks with configurable effort levels.
• Leverage built‑in function calling, web browsing, and Python execution.
• Fine‑tune the model for domain‑specific applications.
• Deploy a compact 20B model for low‑latency inference on limited hardware.
• Experiment with agentic workflows and structured outputs.

When to choose

• You need an open‑source LLM with permissive licensing for commercial or research use.
• Your deployment target is a single high‑memory GPU (80 GB) or a modest 16 GB GPU for the smaller model.
• You want native support for tool use, chain‑of‑thought reasoning, and easy fine‑tuning.
• You prefer a model that can be run directly from PyTorch, Triton, or Metal without additional wrappers.

When to avoid

• You require multi‑GPU or distributed inference out of the box.
• Your environment does not support Python or GPU acceleration.
• You need a model with a different architecture or tokenization scheme.
• You prefer a fully managed hosted API rather than self‑hosting.

Facets

library · maturity active

llm-inference agent-framework prompt-engineering rag large-language-models python windows open-weight mxfp4-quantization gpt-oss-120b gpt-oss-20b single-gpu apache-2.0 ai-agents retrieval-augmented-generation linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="openai/gpt-oss")