NVIDIA says OpenAI's GPT-6 Astra Ultrafast runs on Blackwell, generating tokens up to 8x faster

NVIDIA says OpenAI's GPT-6 Astra Ultrafast runs on NVIDIA Blackwell GPUs and is available now through the OpenAI API and to eligible ChatGPT Work and Codex users.

The headline claim in NVIDIA's blog post is speed: Ultrafast offers up to 8x faster token generation than the Astra Standard mode. NVIDIA attributes that gain to inference optimizations inside OpenAI's models that tap into the capabilities of the NVIDIA Blackwell architecture.

That pairing matters because it ties OpenAI's fastest serving tier directly to NVIDIA's current data-center GPU generation. Rather than treating the GPU as generic plumbing, the post presents Blackwell as the reason the speed-up is possible at all.

For developers, the rollout through the OpenAI API is the practical part of the story: the faster mode is a switch available to applications already built on OpenAI's platform, alongside eligible ChatGPT Work and Codex accounts.

The signal here is about inference economics. As frontier models grow more expensive to serve, per-token latency and throughput become product features, and hardware vendors increasingly market themselves as the reason a given model feels fast.

What to watch next is whether the 8x figure holds up in independent testing and across real workloads, whether Ultrafast becomes the default tier for API customers, and how rival accelerator makers respond now that OpenAI has publicly tied its fastest mode to Blackwell.