LongCat-2.0
Flagshipby Meituan
LongCat-2.0 is Meituan's open-source flagship, released June 30, 2026 under an MIT license on GitHub and Hugging Face. It is a 1.6 trillion parameter mixture-of-experts model with dynamic activation of 33 to 56 billion parameters per token, purpose-built for agentic coding (code understanding, generation, and execution in real-world agent workflows). Native 1 million token context, pretrained from scratch on more than 30 trillion tokens across Chinese, English, multilingual, and code data. Meituan reports 59.5 on SWE-Bench Pro (self-reported, ahead of Gemini 3.1 Pro, GPT-5.5, and Claude Opus 4.6 by their measure); independent verification is pending. The model is the first trillion-parameter release trained and served entirely on a 50,000-card domestic Chinese compute cluster, a milestone for building frontier AI without leading-edge Western chips. A preview version had quietly been running on OpenRouter and longcat.ai for weeks before the announcement, ranking among the top three models globally by call volume during that stealth window.
Input Price
Free
per 1M tokens
Output Price
Free
per 1M tokens
Context Window
1M
tokens
Released
2026-06
Open source
Capabilities
Key Strengths
- ✓1.6T total parameters with 33 to 56B active per token
- ✓MIT open source license
- ✓Native 1M token context
- ✓Purpose-built for agentic coding
- ✓Trained end-to-end on domestic Chinese silicon
Best For
- ▸Self-hosted agentic coding pipelines
- ▸Long-context repository refactors
- ▸Sovereign or geopolitically sensitive deployments
- ▸Research on trillion-parameter MoE training on non-Western hardware
Open Source Model
LongCat-2.0 is free to download and self-host under the MIT. Hosted API pricing varies by provider (e.g., Together, Fireworks, Groq). See our open source LLM guide for deployment options.