LongCat-Flash-Thinking: benchmark results
Provider: Meituan. Released 2025-09-21. Access: Open.
Aggregate Rating 1626 ± 26, rank #367 of 1635 rated models, from 23 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ZeroEval MATH-500 | 99.2 | MATH-500 Score | 100 |
| LLM Stats (AIME 2024) | 93.3 | Score (%) | 95.5 |
| LLM Stats (ZebraLogic) | 95.5 | Score (%) | 87.5 |
| RecRM-Bench - Factual Consistency | 64.98 | Accuracy (%) of the factual-consistency judgement (item and | 83.3 |
| RecRM-Bench - Query-Item Relevance | 75.97 | Accuracy (%) of the three-level relevance score (irrelevant, | 80 |
| RecRM-Bench - Instruction Following | 43.75 | Accuracy (%) of the overall instruction-following compliance | 66.7 |
| ZeroEval GPQA Diamond | 81.5 | GPQA Diamond Score | 63.2 |
| LLM Stats Score | 28.07 | LLM Stats Score (conservative rating) | 59.1 |
| SuperCLUE General (September 2025) - Precise Instruction Following | 41.98 | Score | 50 |
| LLM Stats (MMLU-Redux) | 89.3 | Score (%) | 43.4 |
| SuperCLUE General (November 2025) - Hallucination Control | 77.79 | Score | 32.3 |
| SuperCLUE General (November 2025) - Precise Instruction Following | 26.45 | Score | 27.4 |
Interactive version: theaggregate.ai/model?slug=longcat-flash-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.