Overview

• OpenAI has released gpt-oss-120b and gpt-oss-20b under an Apache 2.0 license, marking its first open-weight models since GPT-2
• Both models are available for free download on Hugging Face, GitHub and major cloud platforms, supporting offline inference without sending data to external servers
• gpt-oss-120b delivers reasoning performance on par with OpenAI’s o4-mini on a single 80 GB GPU; gpt-oss-20b matches o3-mini benchmarks on devices with just 16 GB of memory
• The release is designed to democratize AI by broadening access for developers and organizations and fostering new research and product innovation
• OpenAI partnered with Nvidia, AMD, Cerebras and Groq to optimize hardware compatibility and led extensive security evaluations, including filtering sensitive data and simulating adversarial fine-tuning to mitigate misuse