LongCat-Flash-Chat: benchmark results
Meituan chat model with 560B total parameters and dynamic expert activation: 18.6B–31.3B per token, averaging approximately 27B. Published context is 128K. Provider: Meituan. Released 2025-08-29. Access: Open.
Aggregate Rating 1640 ± 15, rank #331 of 1656 rated models, from 82 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AJ-Bench (LLM-as-a-Judge) - Deep Search | 81.8 | F1 (%) on Deep search (Mind2Web 2-derived items; 52 tasks, 1 | 100 |
| RecRM-Bench - Factual Consistency | 67.34 | Accuracy (%) of the factual-consistency judgement (item and | 100 |
| RecRM-Bench - Instruction Following | 64.84 | Accuracy (%) of the overall instruction-following compliance | 100 |
| BenchTable - Tech | 75.4 | Weighted Score (%) | 79.7 |
| Chatbot Arena (Text - Coding) | 1472 | Arena Score | 70.3 |
| AgentNoiseBench (tau2-bench, Tool Noise) | 48 | Success on a stratified 25 percent sample of tau2-bench task | 69.6 |
| Chatbot Arena (Text - Business, Management & Financial Ops) | 1418 | Arena Score | 68.2 |
| Chatbot Arena (Text - Medicine & Healthcare) | 1441 | Arena Score | 67.3 |
| Chatbot Arena (Text - Math) | 1415 | Arena Score | 66.7 |
| Chatbot Arena (Text - Software & IT Services) | 1452 | Arena Score | 66.7 |
| OpenCompass Research - IFEval | 90.2 | Score (%) | 66 |
| BenchTable - STEM | 61.5 | Weighted Score (%) | 65.7 |
Interactive version: theaggregate.ai/model?slug=longcat-flash-chat · How It Works · Data refreshed daily, snapshot 2026-10-20.