The GPT-OSS models can be downloaded from multiple official sources, with the primary distribution through Hugging Face where gpt-oss-120b and gpt-oss-20b are available. Users can download the model weights directly from the Hugging Face Hub using the Hugging Face CLI with commands like “huggingface-cli download openai/gpt-oss-120b --include ‘original/*’ --local-dir gpt-oss-120b/” for the larger model and similar commands for the smaller variant.
For users who prefer package management, the models are available through PyPI as the gpt-oss package, with different installation options depending on the intended implementation: “pip install gpt-oss” for basic tools, "pip install gpt-oss[torch]" for the PyTorch implementation, "pip install gpt-oss[triton]" for the optimized Triton implementation, and "pip install gpt-oss[metal]" for Apple Silicon hardware. The GitHub repository at github.com/openai/gpt-oss provides the complete source code, reference implementations, and comprehensive documentation for local setup and modification.
Several consumer-friendly platforms also provide easy access to the models: Ollama users can run “ollama pull gpt-oss:20b” or “ollama pull gpt-oss:120b” after installing Ollama, while LM Studio users can download them with “lms get openai/gpt-oss-20b” or "lms get openai/gpt-oss-120b". For production deployments, vLLM provides optimized serving with specific installation commands that include pre-compiled wheels for better performance. Major cloud providers including Microsoft Azure, Amazon Web Services, and various GPU cloud services also offer hosted versions of the models, allowing users to deploy them without managing the underlying infrastructure. The models are also integrated into popular development tools and platforms, making them accessible through familiar interfaces for developers who prefer not to handle direct model downloads and setup.
For more detailed information, see: GPT-oss vs o4-mini: Edge-Ready, On-Par Performance — Dependable, Not Mind-Blowing