Ted Hisokawa
Aug 05, 2025 17:48
OpenAI and NVIDIA have launched new AI models optimized for NVIDIA’s infrastructure, promising advancements in AI development and application across various industries.
OpenAI and NVIDIA have announced the release of two new open-weight AI reasoning models, gpt-oss-120b and gpt-oss-20b. These models aim to make advanced AI development accessible to developers, enterprises, and governments worldwide. According to NVIDIA Newsroom, this collaboration highlights the potential of community-driven innovation and NVIDIA’s key role in democratizing AI technology.
Revolutionizing AI Infrastructure
The new models are designed to be utilized in various sectors such as healthcare, manufacturing, and generative AI applications. OpenAI’s models, trained on NVIDIA H100 GPUs, perform optimally on NVIDIA’s CUDA platform, which is supported by millions of GPUs globally. With these advancements, NVIDIA’s Blackwell architecture offers unprecedented efficiency, achieving 1.5 million tokens per second on the NVIDIA GB200 NVL72 systems.
NVIDIA Blackwell: A Game Changer
NVIDIA’s Blackwell architecture is pivotal in supporting the exponential growth of advanced reasoning models like gpt-oss. It features innovations such as NVFP4 4-bit precision, allowing high-accuracy inference with reduced power and memory requirements. This advancement enables real-time deployment of trillion-parameter large language models (LLMs), unlocking significant economic value for organizations.
Global Access and Development Opportunities
NVIDIA CUDA, a widely accessible computing infrastructure, allows users to deploy AI models across various platforms, including NVIDIA DGX Cloud and GeForce RTX-powered PCs. With over 450 million CUDA downloads, developers can now access these latest models, optimized for NVIDIA’s technology stack. OpenAI and NVIDIA have also partnered with leading open framework providers to offer model optimizations, facilitating diverse development opportunities.
A Legacy of Innovation
This collaboration builds on the longstanding relationship between NVIDIA and OpenAI, which began with the delivery of the first NVIDIA DGX-1 AI supercomputer in 2016. Since then, both companies have worked together to advance AI capabilities, leveraging NVIDIA’s comprehensive software stack to optimize OpenAI’s gpt-oss models.
By enhancing these models for NVIDIA Blackwell and RTX GPUs, NVIDIA aims to accelerate AI development for its extensive community of developers across the globe.
Image source: Shutterstock
