Frontier Log

Benchmark

Humanity's Last Exam

020406080Oct 202657.2%GPT-6 Astra67.7%Opus 5.5020406080Oct 202657.2%GPT-6 Astra67.7%Opus 5.5

Results

ModelScoreSourceHuman-verified
Claude Opus 5.522 Sep 2026 · Anthropic67.7%Introducing Claude Opus 5.5max effortYes
GPT-6 Astra3 Sep 2026 · OpenAI57.2%GPT-6 Astra: A new generation of intelligenceYes

Benchmark

lastexam.aiHigher is better

All benchmarks