OpenAI has upgraded its coding application Codex, which now fully operates on the latest GPT-5.5 model. Deployed on NVIDIA GB200 NVL72 rack systems, this advanced model significantly accelerates developer workflows and expands the capabilities of AI agents in processing information, solving complex problems, and generating new ideas. Over 10,000 NVIDIA employees from various departments, including engineering, product development, legal, and marketing, are already actively using GPT-5.5-powered Codex, describing the results as “stunning” and “life-changing,” highlighting the significant effectiveness of the new technology combination.
The adoption of NVIDIA GB200 NVL72 systems demonstrates significant economic and performance advantages. The cost per million tokens has decreased by 35 times, and token throughput per megawatt has increased by 50 times compared to previous-generation systems. These metrics make the deployment of advanced models economically viable at the scale of large enterprises, opening new horizons for the widespread use of AI agents. In practice, this has led to measurable improvements: debugging cycles that previously took days are now reduced to hours, and experiments that required weeks transform into overnight progress when working with complex multi-file codebases. Teams are now able to release end-to-end features based on natural language prompts with increased reliability and fewer inefficient cycles.
Special attention was paid to enterprise security during the deployment of the Codex application. The system supports secure remote SSH connections to authorized cloud virtual machines, allowing AI agents to work with real corporate data without external exposure. To ensure maximum security and audibility, NVIDIA's IT department has deployed separate cloud virtual machines for each employee. This architecture provides a dedicated isolated environment for the agent, ensuring its operation at maximum capacity with full audibility, while users manage the agent from a familiar interface. NVIDIA's deployment is strictly governed by a zero-data retention policy, and agents access production systems with read-only permissions through command-line interfaces and specialized tools (Skills).
The launch of GPT-5.5 and the implementation of Codex are the culmination of over a decade of strategic collaboration between NVIDIA and OpenAI. This partnership began in 2016 when NVIDIA founder and CEO Jensen Huang personally delivered the first NVIDIA DGX-1 supercomputer to OpenAI's headquarters in San Francisco. Since then, the companies have closely collaborated across all levels of the AI stack. OpenAI has committed to deploying over 10 gigawatts of NVIDIA systems for its next-generation AI infrastructure, laying the groundwork for millions of NVIDIA GPUs to be used in OpenAI's model training and inference for many years to come. The companies are also actively involved in early hardware development and co-design.
Sources
Replies (0)
No replies in this topic yet.
