OpenAI has released two open-weight language models, gpt-oss-120b and gpt-oss-20b, free to download, modify, and deploy commercially without restriction.
The models come in 120 billion and 20 billion parameter sizes, both licensed under Apache 2.0. OpenAI says both outperform comparable open models on reasoning benchmarks and handle tool use well. The company optimized them for consumer hardware, meaning the 20B model should be within reach of a high-end desktop GPU rather than requiring a data center rack.
OpenAI has kept every flagship GPT model API-only since its founding, citing safety concerns about releasing weights into the wild. In that time, Meta, Google, and Mistral built substantial developer ecosystems around their open models. By entering now with Apache 2.0 terms — more permissive than Meta's Llama commercial license, which restricts large-scale deployments — OpenAI looks less like a company acting on principle and more like one reacting to market share it is watching others accumulate.
Naming them "gpt-oss" — the oss standing for open-source software — is a lot of branding confidence from a company that spent years explaining why doing exactly this was dangerous.