Meituan, the Chinese delivery platform giant, open-sourced LongCat-2.0, a 1.6-trillion-parameter coding model that has dominated developer rankings on OpenRouter under the anonymous alias "Owl Alpha" for the past two months. The company revealed the model on GitHub, Hugging Face, and its native platform after weeks of stealth operation.
LongCat-2.0 is a Mixture-of-Experts system built entirely on Chinese semiconductor infrastructure, sidestepping U.S. chip dependencies. The model supports a native 1-million-token context window, enabling it to process substantially longer code files and documentation in single requests compared to typical models capped at 100,000 to 200,000 tokens. This extended context matters for enterprise engineering workflows where developers need to reason across entire codebases.
The release carries a permissive commercial license, positioning it as a direct challenge to closed-source competitors like OpenAI's GPT-4 Turbo and Anthropic's Claude. By open-sourcing an agentic coding model with near-frontier performance, Meituan democratizes autonomous software engineering capabilities that previously remained locked behind expensive API paywalls.
The stealth launch strategy worked. Owl Alpha accumulated sufficient user engagement on OpenRouter to claim top rankings in coding benchmarks before the company attached its own branding. This suggests developers found the model's performance compelling without awareness of its origin or sponsorship.
Training entirely on Chinese chips represents a geopolitical shift in AI development. It demonstrates that frontier-capable models no longer require exclusive access to Nvidia's GPUs or other U.S.-controlled hardware. Chinese infrastructure can produce competitive large language models at scale, reducing Western technological leverage in enterprise AI markets.
LongCat-2.0 specifically targets developers building autonomous agents. The