Model Release
NVIDIA Powers OpenAI's GPT-5.5 with GB200 NVL72 Systems
NVIDIA's GB200 NVL72 systems run OpenAI's GPT-5.5 model, enabling faster and more efficient AI agent operations for enterprise use.
Image: NVIDIA
NVIDIA's GB200 NVL72 rack-scale systems are now powering OpenAI's GPT-5.5 model, which drives the agentic coding application Codex. The system delivers 35x lower cost per million tokens and 50x higher token output per second per megawatt compared to prior-generation systems, making large-scale AI inference more viable for enterprises. Over 10,000 NVIDIA employees across various departments are using GPT-5.5-powered Codex to achieve results they describe as 'mind-blowing' and 'life-changing.' NVIDIA engineers have been testing the model for weeks, with measurable improvements in debugging cycles and experimentation times. Debugging that once took days is now closing in hours, while complex codebases are being processed in overnight cycles. Teams are now able to ship end-to-end features from natural-language prompts with stronger reliability and fewer wasted cycles than earlier models. The collaboration between NVIDIA and OpenAI began in 2016, with NVIDIA delivering its first DGX-1 AI supercomputer to OpenAI's San Francisco headquarters. Since then, the two companies have worked together across the full AI stack, with NVIDIA being a day-zero partner for OpenAI's gpt-oss open-weight model launch. OpenAI has committed to deploying over 10 gigawatts of NVIDIA systems for its next-generation AI infrastructure, which will support millions of NVIDIA GPUs for years to come. *Source: [nvidia](https://blogs.nvidia.com/blog/openai-codex-gpt-5-5-ai-agents/)*
Key points
- NVIDIA's GB200 NVL72 systems run OpenAI's GPT-5.5 model.
- GB200 NVL72 systems deliver 35x lower cost per million tokens and 50x higher token output per second per megawatt compared to prior-generation systems.
- Over 10,000 NVIDIA employees are using GPT-5.5-powered Codex to achieve 'mind-blowing' and 'life-changing' results.
- NVIDIA engineers have been using GPT-5.5 through Codex for weeks, with measurable improvements in debugging cycles and experimentation times.
- Debugging cycles that once stretched across days are closing in hours.
- Teams are shipping end-to-end features from natural-language prompts with stronger reliability and fewer wasted cycles than earlier models.
- NVIDIA and OpenAI have collaborated for over 10 years, starting with the delivery of the first NVIDIA DGX-1 AI supercomputer to OpenAI's San Francisco headquarters.