LongCat-2.0
1.6T-parameter MoE model, ~48B active, with 1M context, trained and served on AI ASIC superpods.
1.6万亿总参大模型，训练全程由国产芯片完成
Official source
Introducing LongCat-2.0longcat.chat · published
Follow-ups
Stealth previews
Benchmarks
| Benchmark | Score | Source check |
|---|---|---|
| SWE-bench ProClaude Code harness · Run on the lab's own corrected version of the public set | 59.5% | Source checkedA person checked this number against its source on 23 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.2 other sourcesThe same number is printed in 2 other lab documents, linked below. |
| GPQA Diamond | 88.9% | Source checkedA person checked this number against its source on 23 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.2 other sourcesThe same number is printed in 2 other lab documents, linked below. |
| BrowseCompin-house harness; context management not stated | 79.9% | Source checkedA person checked this number against its source on 23 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.2 other sourcesThe same number is printed in 2 other lab documents, linked below. |
| Terminal-Bench 2.1Claude Code harness; 6-hour agent timeout | 70.8% | Source checkedA person checked this number against its source on 23 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.2 other sourcesThe same number is printed in 2 other lab documents, linked below. |
Reported by Meituan
- Introducing LongCat-2.0longcat.chat · 30 Jun 2026
Our capture of the source