LongCat-Flash-Thinking
MeituanOpen WeightMIT · Commercial OK
Description
LongCat-Flash-Thinking is Meituan's reasoning model built on the LongCat-Flash foundation with 560B total parameters (MoE, ~27B activated). It introduces a training pipeline specifically tuned for advanced reasoning, featuring Re-thinking Mode that delivers parallel reasoning paths for sophisticated decision-making. Achieves strong performance on mathematical reasoning, agentic tool use, and formal theorem proving benchmarks.
Release Date
2025-09-22
Parameters
560.0B
Context Length
—
Modalities
text
Capability Radar
80
general
70
coding
90
reasoning
68
scienceest.
70
agents
0
multimodal
Science uses a reasoning proxy when dedicated science benchmarks are unavailable.
Rankings
| Domain | #Rank | Score | Source |
|---|---|---|---|
| Agents & Tools | 8 | 74.0 | LS |
| Reasoning | 59 | 62.0 | LS |
Benchmark Scores (LLM Stats)
Agents
BFCL-v3
74.4%SR
Biology
GPQA
81.5%SR
Code
LiveCodeBench
79.4%SR
SWE-Bench Verified
59.4%SR
Communication
Tau2 Telecom
83.1%SR
Tau2 Retail
71.5%SR
Tau2 Airline
67.5%SR
Finance
MMLU-Pro
82.6%SR
General
MMLU-Redux
89.3%SR
Math
MATH-500
99.2%SR
AIME 2024
93.3%SR
AIME 2025
90.6%SR
Reasoning
ZebraLogic
95.5%SR
ARC-AGI
50.3%SR
AA Evaluation Indices
No AA evaluation data available
LLM Stats Category Scores
Language90
Math90
Biology80
Chemistry80
Finance80
General80
Healthcare80
Legal80
Physics80
Reasoning80
Structured Output70
Tool Calling70
Agents70
Code70
Communication70
Frontend Development60
Spatial Reasoning50
Vision50
Pricing
No pricing data available
Speed
No speed data available
Available Providers
(LS internal units)No provider data available