Meituan Releases LongCat-2.5 Large Model, Topping Multiple Agent Benchmarks
Meituan has launched its new-generation large model LongCat-2.5, which ranks first on multiple of 9 evaluation benchmarks across three categories: coding, general tasks, and visual Agent. The model is a native multimodal model designed for long-horizon agent tasks, with 1.6 trillion total parameters and approximately 48 billion parameters activated per token. Building on LongCat-2.0, it strengthens native visual understanding and agentic coding capabilities. It scored 51.8 on AutomationBench, surpassing Claude-Opus-5's 50.3, and also ranked first on BrowseComp with a score of 91.9; on 3DCodeBench it scored 55.4 on text-to-3D. It still lags behind leading overseas models on coding and terminal instruction tasks.
Why read this
AI score 70
The original text lists the specific score gaps between LongCat-2.5 and leading overseas models on 9 Agent benchmarks, which can be used to determine on which tasks the domestic model has caught up.
Topics
Body · AI translation
Meituan Releases New-Generation Large Model LongCat-2.5: Domestic Model, Domestic Chips, Tops Multiple Metrics
Fast Technology Download Client | Exclusive First Look November 03, 2026 07:03:45 From Henan Updated November 03, 2026 07:11:58
Fast Technology News, November 3 — Meituan yesterday launched its new-generation large model LongCat-2.5, which achieved strong results on multiple intelligent Agent test leaderboards, with several items directly surging to first place.
This round of testing covered three major categories—code, general tasks, and visual Agent—across a total of 9 different evaluation benchmarks. The competitors compared were all popular overseas models, including Claude-Opus-5, GPT-5.6-Sol, GPT-5.5, and Claude-Opus-4.8.
On leaderboards including AutomationBench, BrowseComp, LongCatClawBench, and 3DCodeBench, LongCat-2.5 led in score.
AutomationBench scored 51.8 points, surpassing Claude-Opus-5's 50.3 points. BrowseComp scored 91.9 points, also ranking first. In the 3DCodeBench text-to-3D project, LongCat-2.5 scored 55.4, slightly ahead of other competitors.
Simply put, on Agent tasks such as web browsing, automated operations, and 3D generation, LongCat-2.5 has clear advantages. In code and terminal instruction tasks, there is still a slight gap with top overseas models, but overall it has already reached the first tier.
It is understood that LongCat-2.5 is a native multimodal model oriented toward long-horizon agent tasks, with a total parameter count of 1.6 trillion and approximately 48 billion parameters activated per token. Building on LongCat-2.0, it further strengthens native visual understanding and agentic coding capabilities.
Meituan added more long-horizon trajectory data to long-sequence training and, through large-scale long-horizon reinforcement learning, improved the model's end-to-end engineering capabilities in tasks such as complex codebase development and complete project construction.
At the same time, LongCat-2.5 represents visual signals as multi-level discrete tokens, enabling the backbone model to process visual and text inputs in a unified token form.
The combination of native visual understanding and agentic coding capabilities gives the model outstanding performance on Visual Coding and supports tasks such as creative design and 3D.
These upgrades are supported by continuously evolving infrastructure: we optimized long-context training and W4A4 inference on domestic compute, and through training-inference numerical consistency and long-context rollout optimization, supported the efficient operation of large-scale long-horizon RL.
Domestic large models are now intensely competitive, and the results LongCat-2.5 has delivered this time prove that domestic Agent large models already have the strength to compete on the same stage with top overseas products.
"Special Statement: The above work content (including videos, pictures, and audio if any) is uploaded and published by users of 'Dafeng Hao,' a self-media platform under Phoenix Net. This platform only provides information storage space services.
Notice: The content above (including the videos, pictures and audios if any) is uploaded and posted by the user of Dafeng Hao, which is a social media platform and merely provides information storage space services."