Overview
- GPT-OSS-120B features 117 billion parameters and runs on a single NVIDIA GPU while GPT-OSS-20B has 21 billion parameters and operates on laptops with 16 GB of memory.
- Both text-only models integrate chain-of-thought reasoning by calling external tools such as web search and Python code execution.
- OpenAI has published the models on platforms including Hugging Face, AWS, Azure and Databricks under Apache 2.0 licensing to allow free commercial modification.
- Extensive safety evaluations by OpenAI and third-party experts found neither model reaches the company’s “high capability” danger threshold, even after adversarial fine-tuning.
- The open-weight release sets a new foundation for AI innovation ahead of the GPT-5 launch and responds to growing global competition in open-model development.