Meituan LongCat
Meituan's open-weight 1.6T LongCat models and low-cost API
Overview
Meituan LongCat is a family of open-weight models; LongCat-2.0 is a 1.6T mixture-of-experts model (48B active) with a 1M context, MIT-licensed on Hugging Face, and LongCat-2.5-Preview adds image understanding via API. Free to self-host; API $0.30 per million input tokens (limited-time rate).
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 1 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- LongCat-2.0 with FP8 and INT8 variants
- Earlier LongCat models
- LongCat-2.0 and 2.5-Preview
- Cached input $0.006/M
- Volume discounts up to 40%
- OpenAI and Anthropic formats
What you can produce with Meituan LongCat
- LongCat-2.0 1.6T open-weight model
- LongCat-2.5-Preview with image understanding (API)
- 1M-token context with 128K output
- API compatible with OpenAI and Anthropic formats
- Chat at longcat.ai
LongCat-2.0 is one of the most capable open-weight models available, with a million-token context and a permissive MIT licence, and it is notable for being trained without Nvidia hardware. Its benchmark results are Meituan's own, putting it near but behind the latest Claude Opus models on coding, and the API rate is described as a limited-time discount. Watch for the unofficial lookalike site, and use longcat.ai or longcat.chat. A good option for developers wanting a large open model or a cheap long-context API; self-hosting requires a serious GPU cluster.
What we found
LongCat is the family of large language models from Meituan, the Chinese food-delivery and local-services company, which released its first model, LongCat-Flash-Chat, in August 2025. LongCat-2.0, announced on 30 June 2026, is a mixture-of-experts model with 1.6 trillion parameters, about 48 billion of them active for each token, a one-million-token context and up to 128,000 tokens of output, pretrained on more than 35 trillion tokens. Meituan says it was trained entirely on clusters of more than 50,000 AI chips rather than Nvidia GPUs; press reports identify the hardware as Huawei's and call it the first model of its scale trained that way, which Meituan has not confirmed. Its LongCat Sparse Attention builds on DeepSeek's design. Meituan's own benchmarks put it ahead of GPT-5.5 and Gemini 3.1 Pro on SWE-bench Pro but behind Claude Opus 4.7 and 4.8; these are in-house results. Before release it ran anonymously on OpenRouter as "Owl Alpha", which Meituan says reached the top three by call volume.
The LongCat-2.0 weights, with FP8 and INT8 versions, have been on Hugging Face under the MIT licence since early July 2026, as have most earlier LongCat models. On 25 September Meituan released LongCat-2.5-Preview, which adds image understanding and is so far available only through the API. The models run on GPUs and NPUs through SGLang, and the API accepts both OpenAI and Anthropic formats.
The official chat site is longcat.ai and the developer platform is longcat.chat; a similarly named site, longcatai.org, does not appear to be Meituan's. The service is operated by Beijing Kuxun Interactive Technology, a Meituan affiliate.
Self-hosting is free. Meituan's paid API, live since 30 June 2026, charges a limited-time rate of $0.30 per million input tokens and $1.20 per million output for both 2.0 and 2.5-Preview, with cached input much cheaper and volume discounts of up to 40%; OpenRouter charges the same. Mainland China users must verify their identity before paying.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Meituan LongCat's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Meituan LongCat unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Meituan LongCat directly →
Spotted something out of date? Suggest an update →
Alternatives to Meituan LongCat
More in LLMs & Chat
