OpenAI and NVIDIA Unleash Open-Weight AI Models for Global Innovation

OpenAI and NVIDIA have released two new open-weight AI reasoning models — gpt-oss-120B and gpt-oss-20B — optimized to run on the world’s largest AI inference infrastructure. Trained on NVIDIA H100 GPUs, these models are designed for developers, enterprises, and governments to accelerate innovation in generative and reasoning AI across industries.

With support for NVIDIA’s CUDA platform, the models are accessible to over 6.5 million developers worldwide. They are available as NVIDIA NIM microservices, ensuring flexibility, data privacy, and enterprise-grade deployment.

NVIDIA's new Blackwell architecture powers unprecedented performance, achieving 1.5 million tokens per second on a single GB200 NVL72 system. Innovations like NVFP4 4-bit precision reduce power and memory needs, enabling real-time deployment of trillion-parameter LLMs.

In a strong stance on hardware integrity, NVIDIA confirmed its chips contain no backdoors, spyware, or kill switches, rejecting calls for mandatory hardware-based control mechanisms. Drawing lessons from the failed Clipper Chip initiative, NVIDIA emphasized that security should rely on layered defenses, not built-in vulnerabilities.

Together, OpenAI and NVIDIA reaffirm their long-standing partnership in making powerful AI tools widely accessible — all without compromising on trust or security.

Source: OpenAI and NVIDIA Propel AI Innovation With New Open Models Optimized for the World’s Largest AI Inference Infrastructure | NVIDIA Blog