OpenAI has released two artificial intelligence (AI) models — gpt-oss-20b and gpt-oss-120b — that are free for anyone to use and optimised to run both in the cloud and locally on PCs equipped with NVIDIA RTX GPUs.
This collaboration with NVIDIA broadens access to cutting-edge AI technology, allowing developers, businesses and enthusiasts to leverage high-performance models without barriers.
The new gpt-oss models are described as “open-weight”, which gives users the flexibility to download, modify and utilise them in various applications — from web search and research to document processing and coding assistance. Previously, such advanced models were available only as paid cloud services. Under an Apache 2.0 open-source licence, they now come with minimal restrictions and no licencing fees.
The gpt-oss-20b model is designed to run on widely available GPUs with as little as 16GB of VRAM, making it suitable for standard PCs and workstations, while the larger gpt-oss-120b delivers top-tier performance for even the most demanding AI tasks but requires more powerful, specialised hardware for optimal speed.
NVIDIA has fine-tuned these models for RTX hardware to support up to 256 tokens per second on the GeForce RTX 5090 GPU and enable context lengths of up to 131,072 tokens. This makes them some of the most capable models for handling large documents or complex research tasks locally. They leverage the mixture-of-experts (MoE) architecture to provide advanced “chain-of-thought” reasoning, instruction-following and tool-use capabilities.
The models are available through popular AI tools and frameworks such as Ollama, llama.cpp and Microsoft AI Foundry Local so users can easily get started.
Ollama features a simple interface for chatting with the models, supports uploading files and allows developers to work via command line or SDK for more advanced integration. For developers on Windows, Microsoft AI Foundry Local provides a command-line route to access these models, fully optimised for RTX and CUDA with ongoing enhancements on the way.
“The gpt-oss models let developers everywhere build on that state-of-the-art open-source foundation, strengthening US technology leadership in AI — all on the world’s largest AI compute infrastructure,” said Jensen Huang, Founder and CEO of NVIDIA.
Entelechy Asia 