OpenAI Releases Two Open-Weight AI Reasoning Models
- OpenAI and NVIDIA launched two open-weight large language models, gpt-oss-120b and gpt-oss-20b, optimized for NVIDIA’s Blackwell platform.
- The models are designed for high-performance generative, reasoning, and agent-based AI applications and are available globally to developers and enterprises.
- gpt-oss-120b can run on a single NVIDIA 80GB chip, while gpt-oss-20b is suitable for consumer-grade hardware, broadening accessibility.
- The models use a mixture-of-experts architecture and support chain-of-thought reasoning, instruction following, and tool use.
- OpenAI’s move reflects an effort to balance open access with proprietary protections, responding to increased competition in open-source AI.
OpenAI, in collaboration with NVIDIA, has released two new open-weight large language models — gpt-oss-120b and gpt-oss-20b — aimed at advancing reasoning capabilities for developers and enterprises worldwide. The models, optimized for NVIDIA’s Blackwell platform, support a wide range of generative and agentic AI uses.
Details of the New Models and Partnership
The gpt-oss-120b and gpt-oss-20b models represent OpenAI’s first open-weight models in over four years. Developed with a focus on democratizing access to advanced AI, the models are optimized for the latest NVIDIA GPU hardware, including Blackwell GB200 NVL72 systems. The collaboration allows performance of up to 256 tokens per second on NVIDIA’s GeForce RTX 5090 GPU, facilitating both cloud-based and local AI application deployments [1].
Technical Features and Developer Accessibility
Both models use a mixture-of-experts architecture and offer chain-of-thought reasoning, instruction-following, and agentic tool use. gpt-oss-120b, with its high capacity, is designed to run on enterprise-level NVIDIA chips, while the lighter gpt-oss-20b is suitable for consumer devices like laptops. Developers can access and deploy these models through common AI development tools and platforms [1][4].
OpenAI’s Strategic Shift and Industry Context
OpenAI’s decision to open source these models partially is a strategic response to intensified competition from open-source alternatives, particularly from Chinese organizations like DeepSeek. While weights are released and the models are positioned for open access, some proprietary elements such as routing mechanisms and training details remain undisclosed, following a trend among leading AI providers to protect core intellectual property even as access broadens [2].
Enterprise and Developer Adoption
The models target enterprise workloads and broad developer use cases, with particular optimization for generative AI, automation, and research applications. NVIDIA CEO Jensen Huang stated, 'The gpt-oss models let developers everywhere build on that state-of-the-art open-source foundation, strengthening U.S. technology leadership in AI' [1].
Companies mentioned
NVIDIA Corporation stands as a prominent provider of advanced graphics, computational, and networking solutions, operating across the United States, Taiwan, China, and numerous international markets. Its Graphics divisi…