Researcher says GPT-6 Astra cracked a month-long theorem overnight and quietly kept working
morqon · x · 2026-09-06
Math researcher jbrukh reports in two posts that the new Astra model (referred to as GPT-6) delivered a striking research breakthrough:
- Earlier, Astra re-derived results that took him days of constant computation — in about 28 minutes, without looking at previous progress.
- Overnight, unattended, it resolved the next portion of his Seymour's Second Neighborhood Conjecture research program and completed a target theorem his group had pursued for about a month. It also found a methodology that makes proofs of the whole theorem family far more tractable, shrinking a multi-page proof from a previous paper to one paragraph.
- His guess: Astra kept accumulating structural observations about conjecture counterexamples until an insight simplified everything. The model deemed the result unremarkable and didn't alert him.
- He believes there's enough for a follow-up paper but will wait to see how far the family can be pushed.
Note: single-source account, not yet peer-reviewed.
More from Models
- Ornith 1.5 35B A3B beats Flint on hallucination and overthinking in user tests — BarberIcy366 · 2026-10-23
- Meta researcher: horrifying infra, not talent, is why OSS can't catch closed frontier — A_K_Nain · 2026-10-23
- TypeSafe's Jev: a "Decision Model" LLM That Returns Floats at $0.042/M Input Tokens — ariG23498 · 2026-10-23
- Microsoft's MAI-Base-2 scaling ladder pegs the L42 run at ~1.8×10^23 FLOPs — teortaxesTex · 2026-10-23
- Local model agentic coding benchmark: 27B matches 125B, tool-calling is the lever — Joralk_ · 2026-10-23
- NVIDIA open-sources PersonaPlex-7B, a full-duplex voice AI model that listens and talks simultaneously — solyarisoftware · 2026-10-23