Nvidia rolls out Codex at scale

Nvidia said more than 10,000 employees now have early access to OpenAI’s Codex, with GPT-5.5 running on Nvidia Blackwell systems. (blogs.nvidia.com) The April 23 post said the rollout spans engineering, product, legal, marketing, finance, sales, human resources, operations and developer programs. Nvidia described Codex as an “agentic coding application” and said employees are using it for software development and knowledge work. (blogs.nvidia.com) Nvidia tied the deployment to its GB200 NVL72 rack-scale systems, saying GPT-5.5 runs on that infrastructure. The company said the setup delivers 35 times lower cost per million output tokens and 50 times higher output token throughput per megawatt than GPT-4 on Hopper. (blogs.nvidia.com) Codex is software that can read a codebase, suggest changes, write patches and help debug programs from plain-language prompts. Nvidia’s examples included engineers shortening debugging from days to hours and turning multi-file experiments from weeks into overnight runs. (blogs.nvidia.com) OpenAI’s pricing moved at the same time. Its help center says Codex switched from per-message pricing to API-style token usage on April 2 for new and existing Plus, Pro, Business and new Enterprise plans, then expanded that update on April 23 to all existing Enterprise plans, including Edu, Health, Gov and Teachers. (help.openai.com) OpenAI’s API pricing page now lists GPT-5.5 at $5 per 1 million input tokens, $0.50 for cached input and $30 for output. The same page lists GPT-5.4 as a cheaper coding model, which gives buyers a clearer price ladder as they decide where to use higher-end agents. (openai.com) The Nvidia deployment shows coding agents moving beyond a pilot inside one of the world’s largest chip companies. It also shows those tools being sold as a package of model access, infrastructure capacity and measurable token economics. (blogs.nvidia.com) (openai.com) That changes who owns the decision. A tool that writes code now also carries rate cards, cached-input discounts, enterprise plan rules and infrastructure claims that procurement, platform and finance teams can compare line by line. (help.openai.com) (openai.com)