Meituan
Meituan

Longcat Flash Chat

Release Date 2025-09-09(Knowledge cutoff 2025-03-31)131K IN · 131K OUTOpen Model(MIT)

LongCat Flash Chat is Meituan's non-thinking language model for conversation, instruction following and tool use, adjusting expert activation and computation to the input context. Meituan reports efficient inference from overlapping communication and computation, but published output speed reflects particular hardware and serving settings, not a guarantee for every API or local server. AIB places document work and automation in the weaker band at the time checked, supporting comparisons on reviewable dialogue and document tasks before complex automation. Its weights use the MIT license, while the full model remains large despite sparse activation. Choose the checkpoint separately from thinking variants or later generations, and test language quality and tool-call success before integrating it into a service.

API cost

$0.2IN$0.8OUT

Output Speed

Fastper 2,000 tokens
107tok/s
First Output 4.4stotal: 23.2s
SourceArtificial AnalysisLast updated

Expert Performance Evaluation

Which AI gives the answers people prefer most?Arena TextBottom 32%#75
1436

More about Arena Text

Can it handle customer service by the rules?TAU2Bottom 37%#51
79.5%

More about TAU2

28.3%1432

More about Long Context Reasoning

Which AI gives answers people prefer most, without false claims?Arena Text FactualityBottom 22%#82
1429

More about Arena Text Factuality

150110.6%16.528.4%1487

More about Arena Coding

1392143143.1%138914161396

More about Arena Writing, Literature & Language

5.8%143463.6%14551440

More about Humanity's Last Exam

Specialized fields

138213941460146714601429

More about Arena Korean

OddBench

Who survives longer in head-to-head snake games?SnakeBenchBottom 41%#180
20

More about SnakeBench

See each evaluation’s detail page for its source and last updateData corrections & removal requests