Atlas

Models

← All models

LongCat-2.0

Meituan · 2026-06-30 · 1.8T parameters

LongCat-2.0 is Meituan's open-weight large-scale MoE language model, marketed as a 1.6T model with about 48B parameters activated per token; its learned checkpoint tensors contain 1,775,560,457,088 parameters after excluding persistent routing-correction buffers. It was pretrained on more than 35 trillion tokens and introduces LongCat Sparse Attention, N-gram embedding expansion, 1M-context training data, and post-training focused on coding, agentic workflows, search, and optional reasoning.

Benchmark scores

BenchmarkScore
AA-LCR58.0
AA-Omniscience Index-22.6
ALE-Bench1212
Artificial Analysis Agentic Index21.8
Artificial Analysis Coding Index45.3
Artificial Analysis Intelligence Index33.5
Artificial Analysis Omniscience Accuracy29.8
Artificial Analysis Omniscience Hallucination Rate74.7
Artificial Analysis Openness Index38.9
BuseyBench SVG2.2
Capability147.6
CritPt2.6
GDPval-AA v21027
GDPval-AA v2 (normalized)26.3
GPQA Diamond78.0
Humanity's Last Exam (Text-Only)32.1
KernelBench Hard — H100 Mean Peak Fraction10.4
KernelBench Hard — RTX PRO 6000 Mean Peak Fraction16.4
SciCode (Artificial Analysis)35.4
Terminal-Bench 2.150.2
𝜏³-Banking12.8
Loading Atlas data…