GPT-6 Astra Ultrafast goes live on NVIDIA Blackwell: up to 8x faster token generation than Astra Standard

OpenAI switched on GPT-6 Astra Ultrafast this week — up to 8x faster token generation than the Astra Standard mode, built on NVIDIA's Blackwell architecture — and it is live now in the OpenAI API, rolling out to eligible ChatGPT Work and Codex users.

#What happened

OpenAI's inference optimizations for the Blackwell architecture are behind the speed jump: Astra Ultrafast delivers up to 8x faster token generation than the standard Astra mode. It is live now in the OpenAI API and rolling out to eligible ChatGPT Work and Codex users. OpenAI inference lead Philippe Tillet has described how deeply the two companies' engineering now overlaps — including work on models writing their own high-performance kernels for NVIDIA silicon.

#Why it matters

The payoff lands in the loops coding agents run constantly: write code, call a tool, check output, repeat. Faster token generation shortens those edit-test-debug cycles and reduces lag between tool calls, where latency compounds fastest. The open questions: how portable this GPU-specific tuning is for teams not on NVIDIA's stack, and what pricing and quota details Work and Codex users get once wider access rolls out.

SOURCES

1. https://aidailypost.com/news/nvidia-ai-infrastructure-powers-openais-gpt — AI Daily Post, "GPT-6 Astra Gets 8x Speed Boost From NVIDIA Blackwell", published October 2, 2026
2. https://aistartupsnews.com/news/openai-s-gpt-6-astra-ultrafast-8x-faster-with-nvidia-blackwell-gpus/ — AI Startups News, "OpenAI's GPT-6 Astra Ultrafast 8x faster with NVIDIA Blackwell GPUs", published October 2, 2026

Verified October 2, 2026