GPT-6 Astra Ultrafast launches on NVIDIA Blackwell with up to 8x faster tokens
nvidia · x · 2026-10-02
NVIDIA's blog details OpenAI's new GPT-6 Astra Ultrafast mode, running on NVIDIA Blackwell GPUs with up to 8x faster token generation than Astra Standard, now available via the OpenAI API and to eligible ChatGPT Work and Codex users.
- The speedup targets agent loops: shorter edit-test-debug cycles for coding agents, less latency between tool calls, more responsive interactive apps.
- OpenAI inference lead Philippe Tillet says OpenAI's models are "exceptionally good at programming Blackwell and Rubin GPUs" thanks to NVIDIA's tooling, turning that knowledge into high-performance kernels.
- Inference optimizations continue post-deployment, so performance keeps improving.
Related event: GPT-6 Astra Ultrafast Launches on NVIDIA Blackwell with 8x Faster Tokens(2 posts)→
More from Infra
- Redditor builds fully local LLM-powered radio site on two DGX Sparks and a 5090 — jwhh91 · 2026-10-02
- VC quip: many neoclouds are closer to 95% than five nines of reliability — saranormous · 2026-10-02
- Report: lenders demand up to 25% collateral from Nvidia as GPU-backed loans wobble — GaryMarcus · 2026-10-02
- Microsoft Backs Snowflake-Led Effort to Standardize Business Metrics for AI — xiaosun86 · 2026-10-02
- Full SGLang config for GLM-5.3-Flash NVFP4 on 4x RTX 6000 Max-Q — TheZachMueller · 2026-10-02
- GLM-5.3-Flash NVFP4 benchmarks show no per-user speedup beyond 8 concurrent requests — TheZachMueller · 2026-10-02