OpenAI and NVIDIA Release New Open Models for AI Innovation

OpenAI and NVIDIA have released two open-weight AI models designed to make high-performance AI development more accessible. The models named GPT-OSS-120B and GPT-OSS-20B are now available to developers, enterprises, startups, and researchers worldwide.

These models support tasks that require advanced reasoning and generation. They can be utilised in various sectors, including healthcare, manufacturing, and public services. OpenAI trained both models using NVIDIA H100 GPUs and optimized them to run efficiently on NVIDIA's latest Blackwell platform.

Key Updates

• gpt-oss-120b reaches 1.5 million tokens per second on the NVIDIA GB200 NVL72 system

• The models are available as NVIDIA NIM microservices for easy deployment on any CUDA-based infrastructure

• Integration support is available across multiple frameworks, including Hugging Face, FlashInfer, llama cpp, Ollama, and vLLM

• Over 450 million CUDA downloads enable wide access for developers in more than 250 countries

• NVIDIA and OpenAI provide flexible, private, and secure options for deploying these models

Why This Matters

The release of open models supports a shift toward more transparent and customizable AI. Developers can now run these models on their own systems, which improves control over data privacy and deployment. The models also help reduce cost and increase speed for inference across applications.

Infrastructure Advantage

NVIDIA Blackwell introduces features such as 4-bit precision that improve efficiency while maintaining high accuracy. This makes it possible to run very large models in real time without excessive power or hardware requirements.

Looking Ahead

This collaboration strengthens open AI development and gives millions of developers access to powerful tools. With wide compatibility and strong performance, these models mark a new phase in how open AI can support both innovation and scale.

Read Full News Here: Source