Meituan releases LongCat-2.5, a 1.6T-parameter multimodal model for agent tasks

Meituan has released LongCat-2.5, according to an X post by the account leo114119 that links Meituan's LongCat blog. The post says Meituan describes it as a native multimodal model built for agent tasks, with 1.6 trillion parameters, and as strong on complex visual coding projects and professional office work such as creative design, 3D tasks and business analysis.

• Active parameters: Meituan's LongCat account, in its own post on November 2, says the model is natively multimodal with 1.6 trillion total parameters and about 48 billion active, built for long-horizon agentic tasks, with stronger visual understanding and agentic coding. (The first poster said 440 billion were active; Meituan's figure is 48 billion.)
• Benchmarks: the poster reads the published scores as above GPT-5.6 Sol level and mostly level with GLM 5.3 and Kimi K3, with a few ahead. These are the poster's reading of Meituan's own scores, not independent tests.
• Launch offer: new users get 10M free tokens and existing users 5M, with cache hits not drawn from the allowance. A 9.9-yuan promotional package is promised.
• Price: the poster says API pricing is half off for a limited time. Other accounts report 2 yuan in and 8 yuan out per million tokens (about $0.30 and $1.20), a 1M-token context, and support for Claude Code, Codex and Hermes. The poster compares the price with DeepSeek Flash's peak and off-peak rates and says it is about 15% of the peak rate; that comparison is the poster's own.