VentureBeat reported on June 29 that Meituan has unveiled LongCat-2.0, an AI model previously known by the codename "Owl Alpha".

The report said Owl Alpha was run anonymously on OpenRouter over the past two months and stayed near the top of developer charts. Meituan released LongCat-2.0 simultaneously via GitHub, Hugging Face and its own platform.

The company said LongCat-2.0 uses a mixture-of-experts (MoE) architecture with 1.6 trillion parameters. It supports a context window that processes 1 million tokens at a time, and it is released under the MIT licence that allows commercial use.

Benchmark results showed it scored 59.5 on SWE-Bench Pro, beating GPT-5.5's 58.6. It scored 70.8 on Terminal-Bench 2.1 and 77.3 on SWE-Bench Multilingual.

Its pricing policy is also aggressive. There is no charge to retrieve content that has already been read.

Its standard API pricing is $0.75 per 1 million input tokens and $2.95 per 1 million output tokens. During a limited-time promotional period, prices are $0.30 for input and $1.20 for output.

Its training infrastructure has also drawn attention. Meituan trained LongCat-2.0 on a cluster built with more than 50,000 domestic ASICs, without Nvidia GPUs. Meituan was founded in 2010 by Xing Wang (왕싱) and has led China's delivery super-app market. It has recently said it would invest billions of dollars in AI and semiconductor technology in response to slowing profitability.