NVIDIA Blackwell GPUs Accelerate OpenAI’s GPT-6 Astra Ultrafast for Up to 8x Faster Token Generation

GPT-6 Astra Ultrafast, running on NVIDIA Blackwell GPUs, is now available in the OpenAI API and to eligible ChatGPT Work and Codex users. By leveraging inference optimizations that tap into the NVIDIA Blackwell architecture, Ultrafast delivers up to 8x faster token generation compared to Astra Standard mode.

Impact on Developer Workflows

For developers, this speed increase shortens coding agents’ edit-test-debug cycles and reduces latency between tool calls. Faster responses are particularly critical in iterative workflows where an agent writes code, uses a tool, checks results, and decides the next step. Ultrafast brings Astra’s capabilities into these time-sensitive loops, making interactive applications feel more responsive.

Continuous Performance Improvements

OpenAI is using its own models to refine inference software on NVIDIA GPUs, taking advantage of the platform’s programmability to test and implement improvements. Philippe Tillet, inference lead at OpenAI, noted that NVIDIA’s tooling enables models to generate high-performance kernels for Blackwell and Rubin GPUs, optimizing for latency, throughput, and cost.

Uday Ruddarraju, chief technology officer of compute at OpenAI, emphasized that this collaboration helps make AI faster and more useful. The programmable NVIDIA platform allows teams to reuse infrastructure across training, inference, and reinforcement learning, improving utilization as models evolve. Developers can access GPT-6 Astra Ultrafast via the API today, with details on pricing and implementation available in the Ultrafast guide.