Frontier Log

Benchmark · InactiveNo longer reported by the major labs since Apr 2026

τ²-bench

Created by Sierra, 9 Jun 2025 · Higher is better · Website

020406080100Jan 202687.8%Opus 4.576.5%LongCat-Flash-Thinking-2601020406080100Jan 202687.8%Opus 4.576.5%LongCat-Flash-Thinking-2601

Results

ModelScoreSource
Claude Opus 4.524 Nov 2025 · Anthropic87.8%Claude Opus 4.5 System Cardprompt addendum to the policy; Opus 4.1 simulates the user
LongCat-Flash-Thinking-260120 Jan 2026 · Meituan76.5%LongCat-Flash-Thinking-2601 model card (Hugging Face)Corroborated The same number is printed in 2 other lab documents: 1 2average of 4 runs; airline environment fixes from the Claude Opus 4.5 reportOur capture

All benchmarks