How NVIDIA GPUs Help Accelerate OpenAI’s GPT-6 Astra Ultrafast
The new mode delivers up to 8x faster token generation than Astra Standard mode for API, ChatGPT Work, and Codex users.
- OpenAI used internal models to generate optimized inference kernels directly for Blackwell and Rubin GPU architectures.
- The acceleration targets low-latency multi-step loops, including tool calls and coding agent edit-test-debug cycles.
- Access is available immediately through the OpenAI API, ChatGPT Work, and Codex.
Developers building agentic workflows and interactive coding assistants get significantly reduced latency between iterative tool calls.
