FrontierMath Tier 4 fully solved: GPT-6 Astra cracks the last problem standing
Jsevillamol · x · 2026-09-11
Epoch AI reports that every FrontierMath Tier 4 problem has now been solved by AI, with GPT-6 Astra solving the last one, authored by Jay Pantone. The benchmark's designer adds: the set was built in the o4-mini era with a slow solve rate expected; his January prediction of saturation within nine months proved accurate despite some invalidated problems.
- He admits he still can't solve most of the problems himself, and probably never will.
- Combined with FirstProof results, the chance of finding a problem with a known answer that no vanilla or harnessed AI model can solve is near zero.
- Mathematicians often noted AI found unintended shortcuts on Tier 4 problems — not so for this last one.
His takeaway: our understanding of what is truly hard in mathematics needs a complete rethink.
More from AGI Musings
- Anthropic's S-1 Shows $6.6M Revenue Per Employee as AI Agents Log 6x Human Hours — jacalulu · 2026-10-26
- Ben Recht: The world needs a truly open language model, and it's easier to build than you think — beenwrekt · 2026-10-26
- Which jobs fall to AI first? The rule: any layer whose rules can be written as code — FinanceYF5 · 2026-10-26
- Anthropic S-1: expert-level AI 'within a few years', 90% of code written by Claude — kimmonismus · 2026-10-26
- The identity crisis in software engineering: multiagent systems demand a new problem set — rakyll · 2026-10-26
- Frontier LLMs negotiate contracts poorly and frequently renege for selfish gain, study finds — xuanalogue · 2026-10-26