BREAKING
Meituan Opens 1.6T LongCat-2.0
0T
Total params
0B
Active params
0M
Context tokens
Trained on Domestic Chips
SWE-bench Pro: 59.5
LongCat-2.059.5
GPT-5.558.6
Open Power, Heavy Hardware
Strengths
MIT license, public code
1M-token context
Agent-native support
Drawbacks
BF16 near 3.55TB
Needs GPU/NPU clusters
Reported loops, confusion
Frontier Coding Goes Open
AI NEWS BLITZ
Meituan has fully open-sourced its 1.6-trillion-parameter LongCat-2.0 model under the MIT license.