Meituan releases LongCat-2.5 Preview, adding native multimodal and long-horizon Agent capabilities

Article

Meituan LongCat-2.5 Preview released: native multimodality, focused on long-horizon Agent Dongcha Beating AI news flash: Meituan has released LongCat-2.5-Preview. The model retains LongCat-2.0's 1.6T total parameters, approximately activated parameters, and token context, while adding native multimodal capabilities and beginning to execute long-horizon tasks across terminals, browsers, desktop software, spreadsheets, design tools, and other scenarios.

LongCat-2.5-Preview has been integrated into the official API, while also supporting OpenAI and Anthropic APIs, with a maximum output length of token s. The official announcement also provides integration methods for Codex, OpenCode, OpenClaw, and CatPaw; existing Agent tools can call it directly.

Meituan has also granted all existing users 5000000 free token s to try the new model, while existing token plans can continue to be used. However, the official announcement has not yet disclosed LongCat-2.5's benchmark，GUI operations and how much long-horizon tasks have improved; no public data is currently available for comparison.

Original link https://m.theblockbeats.info/flash/369079

Key points

Meituan releases LongCat-2.5-Preview.
The model retains LongCat-2.0's 1.6T total parameters, approximately activated parameters, and token context.
The model adds native multimodal capabilities for executing long-horizon tasks across terminals, browsers, desktop software, spreadsheets, and design tools.
The model has been integrated into the official API, supports OpenAI and Anthropic APIs, and has a maximum output length of token s。
The official announcement provides integration methods for Codex, OpenCode, OpenClaw, and CatPaw.
Meituan has granted all existing users 5000000 free token s; existing token plans can continue to be used.
The official announcement has not yet disclosed LongCat-2.5's benchmark，GUI No public data is currently available to compare improvements in operations and long-horizon tasks.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Multimodality means that a model can process information beyond text. Long-horizon Agent tasks mean that the model performs a series of operations in software such as terminals and browsers.

Why it matters to readers

This means existing Agent tools can call the model and use it for the various software-operation scenarios listed in the article.

Beginners can first try the model through the official API or related tools, but its actual performance improvement cannot currently be determined from public information alone.

Risks and unknowns

The official announcement has not yet disclosed benchmark。
GUI No public data is available to compare actual improvements in operations and long-horizon tasks.

Meituan releases LongCat-2.5 Preview, adding native multimodal and long-horizon Agent capabilities

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center