AI brokers have revolutionized developer workflows, and their subsequent frontier is information work: processing data, fixing complicated issues, developing with new concepts and driving innovation.
Codex, OpenAI’s agentic coding utility, is enabling this new frontier. It’s now powered by GPT-5.5, OpenAI’s newest frontier mannequin, which runs on NVIDIA GB200 NVL72 rack-scale techniques.
Over 10,000 NVIDIANs — throughout engineering, product, authorized, advertising, finance, gross sales, HR, operations and developer applications — are already utilizing GPT-5.5-powered Codex to attain, of their phrases, “mind-blowing” and “life-changing” outcomes.
NVIDIA engineers have had entry to GPT-5.5 by way of the Codex app for just a few weeks, and the good points are measurable. Served on GB200 NVL72, which is able to delivering 35x decrease price per million tokens and 50x larger token output per second per megawatt in contrast with prior-generation techniques — economics that make frontier-model inference viable at enterprise scale.
Debugging cycles that after stretched throughout days are closing in hours. Experimentation that beforehand required weeks is popping into in a single day progress in complicated, multi-file codebases. Groups are delivery end-to-end options from natural-language prompts, with stronger reliability and fewer wasted cycles than earlier fashions.
OpenAI’s beautiful progress is simply the newest instance of NVIDIA’s work with each frontier mannequin firm — not simply to speed up the usage of AI brokers inside NVIDIA, however to assist the corporate’s companions construct the world’s greatest, lowest price and most energy environment friendly fashions for everybody.
As NVIDIA founder and CEO Jensen Huang advised workers in a company-wide e-mail urging everybody to make use of Codex: “Let’s leap to lightspeed. Welcome to the age of AI.”
A Deployment Constructed for Enterprise Safety
Similar to people, each agent wants its personal devoted laptop.
To make sure seamless operation inside safe enterprise environments, the Codex app helps distant Safe Shell (SSH) connections to permitted cloud digital machines, permitting brokers to work with actual firm knowledge with out exposing it externally.
So to make sure most safety and auditability, NVIDIA IT rolled out cloud digital machines (VMs) for each worker to run their agent safely. This supplies a devoted sandbox for the agent to function at its most capabilities whereas sustaining full auditability. Customers can management the Codex agent operating within the cloud VM from a consumer interface that each worker is accustomed to.
A zero-data retention coverage governs NVIDIA’s deployment, and brokers entry manufacturing techniques with read-only permissions by way of command-line interfaces and Abilities — the identical agentic toolkit NVIDIA makes use of to run automation workflows throughout the corporate.
A Decade of Full-Stack Collaboration
The GPT-5.5 launch and the Codex rollout mirror greater than 10 years of collaboration between NVIDIA and OpenAI. The partnership started in 2016, when Huang hand-delivered the primary NVIDIA DGX-1 AI supercomputer to OpenAI’s San Francisco headquarters.
Since then, the 2 corporations have labored carefully throughout the total AI stack.
NVIDIA was a day-zero companion for OpenAI’s gpt-oss open-weight mannequin launch, optimizing mannequin weights for NVIDIA TensorRT-LLM and ecosystem frameworks together with vLLM and Ollama.
OpenAI has dedicated to deploying greater than 10 gigawatts of NVIDIA techniques for its next-generation AI infrastructure — a buildout that may put hundreds of thousands of NVIDIA GPUs on the basis of OpenAI’s mannequin coaching and inference for years forward.
And OpenAI and NVIDIA are early silicon and codesign companions: OpenAI supplies suggestions that informs NVIDIA’s {hardware} roadmap, and in flip good points early entry to new architectures. That relationship produced a concrete milestone — the joint bring-up of the primary GB200 NVL72 100,000-GPU cluster. The cluster accomplished a number of large-scale coaching runs and set a brand new benchmark for system-level reliability at frontier scale.
GPT-5.5 is the product of that infrastructure operating at full power.
Study extra in OpenAI’s announcement.
