LongCat-2.5

Meituan multimodal model for long-horizon agentic coding, visual coding, design, and 3D tasks.

Also known as LongCat 2.5

What it does

LongCat-2.5 is Meituan's native multimodal model for long-horizon agentic tasks. The November 2, 2026 news post says it was released that day, that people can try it at longcat.ai, and that developers can call an API that launched the same day. The technical blog says it has 1.6 trillion total parameters and approximately 48 billion activated per token, and that it builds on LongCat-2.0 with stronger native visual understanding and agentic coding. The blog says that combination supports visual coding, creative design, and 3D tasks. The news post says training ran entirely on domestic compute, and that on several mainstream benchmarks the scores sit between Claude Opus 4.8 and Claude Opus 5, with some above Claude Opus 5. That comparison is reported by the news post. The API pricing page labels its pay-as-you-go rates a limited-time discount: uncached input $0.30, cached input $0.006, and output $1.20 per million tokens, and also lists ¥2, ¥0.04, and ¥8. It does not print an undiscounted price.

Who is LongCat-2.5 for?

Developers who want a multimodal model for long coding sessions, visual coding, design, and 3D tasks.

Tags

• multimodal
• coding
• agent