NVIDIA teams up with OpenAI to roll out optimized open-source GPT models for RTX GPUs. The new gpt-oss-20b and gpt-oss-120b models deliver fast, smart inference from cloud to PC—running at up to 256 tokens per second on the GeForce RTX 5090.
These models bring agentic AI tasks like web search and deep research to a wider audience, with flexible chain-of-thought reasoning and adjustable complexity. They support massive context windows — up to 131,072 tokens — making them top candidates for coding help, document analysis, and complex queries. Trained on NVIDIA H100 GPUs, the models use efficient MXFP4 precision to boost quality without hogging resources.
Jensen Huang, NVIDIA CEO, said: 
“OpenAI showed the world what could be built on NVIDIA AI — and now they’re advancing innovation in open-source software.”
“The gpt-oss models let developers everywhere build on that state-of-the-art open-source foundation, strengthening U.S. technology leadership in AI — all on the world’s largest AI compute infrastructure.”
Testing these models is easiest through the new Ollama app, fully optimized for RTX with minimal setup. Quickly chat with the models on GPUs with 24GB+ VRAM, and leverage features like PDF/text file support, multimodal input, and adjustable context limits. Ollama also offers CLI and SDK access for developers building AI-powered apps.
Other integrations include llama.cpp and Microsoft AI Foundry Local on Windows, which utilize NVIDIA CUDA and will soon add TensorRT support for faster inference.
NVIDIA continues pushing RTX GPU performance with open-source contributions to llama.cpp and GGML, adding CUDA Graphs and CPU overhead reductions. The llama.cpp GitHub repo is a good starting point for enthusiasts.
NVIDIA’s push signals a major step in bringing open-source AI reasoning models to personal and professional RTX AI PCs, empowering developers and users to build smarter, faster AI tools right from their desktops.
Stay connected with NVIDIA AI PC on Facebook, Instagram, TikTok, and X, and subscribe to the RTX AI PC newsletter.
Join NVIDIA’s Discord server for developer discussions and community projects around RTX AI innovation.
