NVIDIA reports that Codex, its internal development system built on GPT‑5.5, has accelerated end‑to‑end research workflows by roughly 10x following enterprise deployment, and that roughly 40,000 NVIDIANs now have access across the company. The deployment runs in production on NVIDIA GB200 and GB300 hardware and supports longer, more autonomous sessions that the company says can surface bugs and ideas beyond the scope of the original prompts. If sustained, that scale of speedup could change how engineering and research teams run experiments and deliver production systems.
The company describes Codex as operating in production on GB200 and GB300 accelerators, enabling extended sessions that do more than respond to single, short prompts. According to the report, those longer sessions allow the system to carry context forward, explore failure modes and generate lines of inquiry that engineers might not have opened manually — functions NVIDIA highlights as contributing to faster end‑to‑end progress on research tasks.
Adoption has been broad internally: the company says roughly 40,000 employees have been granted enterprise access, and the coding agents team reports that engineers routinely use Codex as their default tool for complex engineering problems. Dennis Hannusch of the coding agents team is named as the source for the account that engineers lean on Codex to run full ML experiments and to assist in shipping production systems, suggesting the tool is integrated into both exploratory research and delivery pipelines.
Sources
Replies (0)
No replies in this topic yet.
