OpenAI’s GPT-6 Astra Ultrafast is available in its API and to eligible ChatGPT Work and Codex users, with NVIDIA citing up to eightfold faster token generation.
OpenAI’s GPT-6 Astra Ultrafast became available on October 1, 2026, through the OpenAI API and to eligible ChatGPT Work and Codex users, according to NVIDIA. The companies attribute its speed to OpenAI inference optimizations running on NVIDIA Blackwell GPUs. NVIDIA says the mode can generate tokens up to eight times faster than GPT-6 Astra Standard; the article does not provide a benchmark methodology or workload-specific results for that comparison.
NVIDIA positions the lower-latency mode for tasks where a model repeatedly responds, calls tools and continues working. In software development, for example, faster generations could shorten the wait between an agent proposing code, testing it and deciding on a follow-up. That is the practical use case described by the companies, rather than a claim that every application will see the same speedup.
OpenAI inference lead Philippe Tillet said NVIDIA’s tooling and documentation helped OpenAI develop models that can program Blackwell and Rubin GPUs. OpenAI CTO of compute Uday Ruddarraju said the company used its internal models to optimize inference software on NVIDIA GPUs, with NVIDIA’s programmable architecture supporting the acceleration. These comments describe the companies’ account of their joint technical work.
The post also frames performance as ongoing work after deployment: OpenAI says it uses its models to help improve inference software running on NVIDIA hardware. NVIDIA argues that a programmable platform can be reused across training, inference and reinforcement learning as workload needs change. The article does not specify a future optimization schedule or quantify the infrastructure-utilization benefits.
For developers, the immediate change is access to a faster-generation option through the API. NVIDIA directs users to OpenAI’s Ultrafast guide for access, pricing and implementation information, but its own article does not state those terms. Availability in ChatGPT Work and Codex is limited to eligible users, rather than described as a universal rollout.