OpenAI researchers on GPT-6 Astra: original math before launch, 1970s bound improved
a16z · x · 2026-09-08
In an a16z conversation recorded before GPT-6 Astra's launch, OpenAI's Mark Sellke and Mehtaab Sawhney recount: GPT-5 surfaced a published solution to an open Erdős problem in five minutes, then found solutions to 10+ more supposedly open problems; Astra improved a sphere-packing bound that stood since the 1970s (Sawhney made zero progress on it in six months of grad school) and produced a 15-page proof establishing the existence of non-sofic groups.
Related event: OpenAI's Astra Cracks Progress on Ten Open Math Problems(5 posts)→
More from Models
- Watch Astra beat Montezuma's Revenge, RL's infamous benchmark game — emollick · 2026-09-09
- Model Can't Draw ASCII Whales, So It Writes a Node.js Script to Improve — teortaxesTex · 2026-09-09
- Blender head-to-head: same prompt, 12 seconds, and one frontier model is in a different class — ZeroStateReflex · 2026-09-09
- Netlify adds GPT-6 Astra, Gemini 3.8 Flash, Claude Fable 5.1 and smarter Agent Runner scoping — thisiskp_ · 2026-09-09
- GLM-5.3-Flash tops agentic tool-call leaderboard at 78%, priced at just $0.50/M output tokens — shensi · 2026-09-09
- Artificial Analysis Updated Benchmarks Twice in 4 Days for Astra — py-net · 2026-09-09