GPT-6 Astra hits 97.4% on FrontierMath Tier 4, trained on 100k+ GPUs
ChrisGPT · x · 2026-09-04
Ahead of the GPT-6 Astra launch, newly reported details include:
- 97.4% on FrontierMath Tier 4
- VP of research Aidan Clark calls it OpenAI's "by far our largest scale training run," pre-trained on more than 100,000 GPUs
- It's OpenAI's first model where previous AI model generations were heavily involved in training the new one, with older models helping train their successor
AI models helping train the next generation is the most notable signal here.
Related event: GPT-6 Astra Trained on Over 100,000 GPUs in Record Run(4 posts)→
More from Infra
- Neon-pioneered architecture underpins Lakebase, presented at VLDB — matei_zaharia · 2026-09-04
- 12B local models now rival GPT-4, and tiered local AI is coming, argues Wardell — draginol · 2026-09-04
- QTS's own site confirms zero-water cooling at its data centers — GlenBradley · 2026-09-04
- Virginia State Study: Most Data Centers Use No More Water Than a Large Office Building — GlenBradley · 2026-09-04
- Qwen3.8-27B hits ~130 tok/s on Kaggle's free TPU with full 262k context — A-Rahim · 2026-09-04
- Agent cost benchmark: the money goes to retrieval, not reasoning — 66% cheaper with memory — Coworker_ai · 2026-09-04