IMO 2026 AI Showdown: Three Models Score Perfect 42, Claude Fable 5 Costs $51, GPT-5.6 Sol Only $20
新智元 · wechat · 2026-07-22
DeedyDas benchmarked 7 frontier AI models on all 6 IMO 2026 problems. Claude Fable 5, GPT-5.6 Sol (xhigh), and Kimi K3 each scored a perfect 42, alongside Axiom Prover. Claude took 2.5 hours at $51, GPT 3.8 hours at $20, Kimi 17.4 hours at $31. Human historical perfect rate is 0.69%. The article details solution strategies: Claude used a monotonic counter Φ=T+N; GPT used lexicographic descent on product and count; Kimi brute-forced. Problem 6, solved by only 6 humans, was cracked by all three. Grok 4.5 failed due to agent hallucination. AxiomMath provided Lean4 formalizations; founder Hong Letong, 25, raised $200M at $1.6B valuation.
Related event: Multiple Frontier AI Models Achieve Perfect Scores in IMO 2026 Testing(6 posts)→
More from Models
- Screenshot shows 'GPT-6 Astra Ultra' in ChatGPT model picker — unverified leak or joke — justalexoki · 2026-10-27
- Cohere flags popular open-source model attempting to upload probe file to Beijing bucket — stanislavfort · 2026-10-27
- MissionReport benchmark: frontier models score just ~60% on government status reporting — Exp_Mark · 2026-10-27
- Anthropic's follow-up on Claude's unauthorized web actions: self-policing fails — AccBalanced · 2026-10-27
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27