# LongCat 2.0

> **TL;DR.** LongCat-2.

- **Category:** AI / Foundation Models / Agentic Coding
- **Stage:** validating
- **Age:** 76 days
- **Origin date:** 2026-04-20
- **First detected:** 2026-07-05
- **Canonical URL:** https://earlyterms.com/term/longcat-2-0
- **Sources:** 7 primary URLs

## Definition

LongCat-2.0 is a 1.6-trillion-parameter mixture-of-experts language model from Meituan built for agentic coding, with 1M-token context and roughly 48B parameters active per token. It is the first model of this scale trained end-to-end — pretraining and inference — entirely on domestic Chinese AI chips rather than Nvidia GPUs.

Meituan quietly began testing the model in April 2026, and it spent two months topping OpenRouter's usage charts anonymously as ['Owl Alpha'](https://decrypt.co/372579/longcat-2-0-meituan-ai-stealth-model-openrouter) before the company confirmed the connection and open-sourced LongCat-2.0 under MIT license on [June 30, 2026](https://venturebeat.com/technology/meituan-open-sources-longcat-2-0-the-1-6t-near-frontier-agentic-coding-model-thats-been-leading-openrouter-trained-entirely-on-chinese-chips), alongside SWE-bench Pro and Terminal-Bench scores rivaling GPT-5.5.

## Analogy

It's a spy who worked undercover for two months before revealing which agency sent them.

## Why it's emerging now

Meituan revealed on June 30, 2026 that the mysterious 'Owl Alpha' topping OpenRouter's charts for two months was actually LongCat-2.0, a 1.6T-parameter model trained entirely on domestic Chinese chips. It scored 59.5 on SWE-bench Pro — ahead of GPT-5.5's 58.6 — while undercutting frontier pricing, triggering renewed debate over whether US export controls still constrain China's AI ambitions.

## Related terms

- *competitor:* DeepSeek-V4-Pro
- *competitor:* Qwen3.6
- *competitor:* GLM-5.2
- *competitor:* Kimi K2.7 Code
- *competitor:* MiniMax M3
- *competitor:* GPT-5.5
- *alias:* Owl Alpha
- *related:* Hermes Agent
- *related:* OpenClaw
- *related:* Claude Code
- *parent:* agentic coding
- *related:* agent harness

## Sources

1. [VentureBeat: Meituan open sources LongCat-2.0](https://venturebeat.com/technology/meituan-open-sources-longcat-2-0-the-1-6t-near-frontier-agentic-coding-model-thats-been-leading-openrouter-trained-entirely-on-chinese-chips)
2. [South China Morning Post: China debuts biggest AI model trained on local chips](https://www.scmp.com/tech/tech-trends/article/3358854/china-debuts-biggest-ai-model-trained-local-chips-meituan-releases-longcat-20)
3. [Decrypt: The Stealth AI Model That Was Quietly Topping OpenRouter All Along](https://decrypt.co/372579/longcat-2-0-meituan-ai-stealth-model-openrouter)
4. [LongCat AI: official LongCat-2.0 model page](https://www.longcatai.org/models/longcat-2)
5. [meituan-longcat/LongCat-2.0 model card (Hugging Face)](https://huggingface.co/meituan-longcat/LongCat-2.0)
6. [meituan-longcat/LongCat-2.0 (GitHub)](https://github.com/meituan-longcat/LongCat-2.0)
7. [Hacker News: LongCat-2.0 launch discussion](https://news.ycombinator.com/item?id=48727116)

---
_Generated by EarlyTerms · https://earlyterms.com/term/longcat-2-0_