OpenAI launches GPT-6 Astra Ultrafast on NVIDIA Blackwell GPUs

OpenAI has launched GPT-6 Astra Ultrafast, a mode running on NVIDIA Blackwell GPUs. It is now available through the API and to eligible ChatGPT Work and Codex users.

OpenAI says Ultrafast delivers up to eight times faster token generation than Astra Standard. The improvement is intended to shorten coding agents’ edit-test-debug cycles and reduce pauses between tool calls.

Optimized for agent workflows

The speed benefit is most relevant to workflows in which an agent writes code, uses a tool, checks the result and chooses its next step. OpenAI says Astra Ultrafast brings Astra’s capabilities to these latency-sensitive loops.

The company is also using its own models to improve the inference software running on NVIDIA GPUs. OpenAI says the platform’s programmability allows it to test and implement those changes, potentially making deployed infrastructure more productive over time.

OpenAI comments on Blackwell

Philippe Tillet, OpenAI’s inference lead, said NVIDIA’s investment in tooling and documentation had helped the company optimize its models for Blackwell and Rubin GPUs. Uday Ruddarraju, OpenAI’s chief technology officer of compute, said OpenAI’s internal models were used to optimize inference on NVIDIA GPUs.

OpenAI says NVIDIA’s programmable platform allows infrastructure to be reused across training, inference and reinforcement learning as models evolve. Access, pricing and implementation details are available in the Ultrafast guide.