Accelerating Engineering and Research
On May 12, 2026, NVIDIA engineers detailed their use of Codex, built on GPT-5.5 and deployed on NVIDIA GB200 and GB300 infrastructure, to enhance both production development and research experiments.
Building and Shipping Production Systems
Members of NVIDIA’s coding agents team have adopted Codex as their primary tool for complex engineering tasks across the organization. Senior Software Engineer Dennis Hannusch stated that Codex with GPT-5.5 “surfaces bugs and gaps” that earlier models overlooked and requires “much less handholding.” He noted that the tool has been instrumental in evolving an internal platform from a minimum viable product (MVP) to a production-ready system, significantly improving scalability and reliability.
The team also developed a podcast recording application, similar to Riverside, in just hours using Codex. Hannusch emphasized that due to privacy constraints, traditional software procurement would have taken weeks; however, the Codex desktop app facilitated autonomous testing of video and audio recording functionalities during development.
Running Full Research Workflows
For NVIDIA’s research teams, Codex has largely automated the research loop, which includes identifying research topics, generating scripts for machine learning experiments, and executing jobs on remote machines. AI researcher Shaunak Joshi described GPT-5.5 as a “creative partner” when the team directed it towards a corpus of relevant papers, stating that it suggested a knowledge graph to help visualize connections among ideas.
Joshi also highlighted that the Codex app supports SSH, allowing researchers to run large machine learning workloads from their laptops without the need for manual host setup. He reported that Codex is effective in translating older codebases, explaining that many are “taking their Python repository, sending it to GPT-5.5, and it’s rewriting it into Rust and making it like 20X more efficient.”
Significant Improvements and Future Plans
NVIDIA has observed a 10x speed improvement in end-to-end research workflows, with approximately 40,000 NVIDIANs currently having access to Codex. The company expressed enthusiasm about the ongoing potential of Codex, with Hannusch stating, “Codex has completely changed the threshold for what’s worth building.” He added that they are just beginning to explore the full capabilities of the tool, aiming to continue moving ideas from concept to execution within a single workflow.
Original source: OpenAI News