OpenAI has released two open-weight reasoning models, gpt-oss-120b and gpt-oss-20b, under the Apache 2.0 license. These text-only models are designed for agentic workflows with strong instruction following, tool use, and adjustable reasoning effort.
The models utilize a Mixture-of-Experts architecture with MXFP4 quantization to reduce memory requirements, allowing the larger model to fit on a single 80GB GPU. OpenAI conducted safety evaluations confirming that neither the default nor adversarially fine-tuned versions reach high capability in biological or cyber risk domains.
The release includes full chain-of-thought outputs and support for Structured Outputs, aiming to provide customizable reasoning capabilities while maintaining safety standards across the open-source ecosystem.