Introducing GPT‑5.5 with NVIDIA researcher
- OpenAI published a video on April 24 featuring NVIDIA researcher Shaunak Joshi describing how GPT‑5.5 sped up his experiments, days after OpenAI formally launched the new model on April 23. - NVIDIA said GPT‑5.5 now powers Codex on GB200 NVL72 systems, and more than 10,000 NVIDIA employees are already using it; one researcher said it cut experiment time by 10x. - The launch extends a decade-long OpenAI-NVIDIA partnership and puts GPT‑5.5 into ChatGPT, Codex, and the API one day after release. (openai.com) (blogs.nvidia.com)
OpenAI’s GPT‑5.5 launch quickly turned into a customer case study: NVIDIA researcher Shaunak Joshi said the model made some of his experiments run 10 times faster. (youtube.com) (openai.com) OpenAI introduced GPT‑5.5 on April 23, 2026, calling it its “smartest and most intuitive” model for coding, research, data analysis, and tool use across software. (openai.com) A day later, on April 24, OpenAI published a video with Joshi saying the “magic moment” came when GPT‑5.5 answered an abstract research question with a workable solution. (youtube.com) The underlying idea is agentic software: instead of answering one prompt at a time, the model plans steps, uses tools, checks its work, and keeps going until a task is finished. (openai.com 1) (openai.com 2) That is the pitch behind Codex, OpenAI’s coding product, which NVIDIA said is now powered by GPT‑5.5 running on NVIDIA GB200 NVL72 rack-scale systems. (blogs.nvidia.com) NVIDIA said more than 10,000 employees across engineering, legal, finance, sales, human resources, and operations are already using GPT‑5.5-powered Codex internally. (blogs.nvidia.com) The company framed the gains in concrete terms. NVIDIA said its engineers had been using GPT‑5.5 in Codex for weeks, and Joshi said tasks that once took days of trial and error could be compressed into far fewer iterations. (blogs.nvidia.com) (youtube.com) OpenAI also stressed speed alongside capability. The company said GPT‑5.5 matches GPT‑5.4 on per-token latency in real-world serving while using fewer tokens on the same Codex tasks. (openai.com) The release widened within 24 hours. OpenAI updated its announcement on April 24 to say GPT‑5.5 and GPT‑5.5 Pro were available in the application programming interface, in addition to ChatGPT and Codex rollouts. (openai.com) NVIDIA tied the launch to a longer relationship, saying its work with OpenAI goes back to 2016, when Jensen Huang delivered the first DGX‑1 system to OpenAI’s San Francisco office. (blogs.nvidia.com) So the video was not just a product testimonial. It showed OpenAI using a top infrastructure partner to argue that GPT‑5.5 is already being used on real workloads, by named researchers, on production NVIDIA hardware. (youtube.com) (blogs.nvidia.com)
