Breaking

Meituan Unveils LongCat-2.5, a 1.6-Trillion-Parameter Multimodal Model Built for Agentic Work

November 3, 2026 at 10:45 EST

Foundation Models
AI Agents
Software Dev & Coding

Chinese food-delivery and local-services giant Meituan has released LongCat-2.5, a natively multimodal model with 1.6 trillion total parameters and roughly 48 billion active parameters, positioned for long-horizon agentic tasks.

Model release · Meituan

Meituan unveils LongCat-2.5, a 1.6-trillion-parameter multimodal model built for long-horizon agent work

The natively multimodal model targets extended, multi-step tasks in visual understanding, agentic coding, and creative and 3D design. Only about 48 billion of its parameters are active for any given computation.

1.6T

Total parameters

~48B

Active parameters per computation

~3%

Share of the model engaged at once

Total vs. active parameters, drawn to scale

Each block represents 80 billion parameters. The total model is 20 blocks; the active share is roughly 0.6 of one block.

1.6T

Total

~48B

Active

Roughly 33× more parameters in total than are active at any moment.

Focus areas

The capabilities are announced; the independent evidence is not yet in.

Stronger visual understanding
Improved agentic coding for complex coding projects
Creative design and 3D tasks

Not yet verified: independent benchmark results, pricing, licensing terms, and availability to outside developers. Claims about how LongCat-2.5 compares with rival models cannot yet be checked.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.