Launching today
LongCat-2.5
A multimodal AI model built for complex agentic tasks
1 follower
A multimodal AI model built for complex agentic tasks
1 follower
LongCat-2.5 is Meituan's 1.6-trillion-parameter native multimodal model, designed for long-horizon agentic coding and complex engineering tasks. It combines visual understanding with multi-step reasoning and tool use to build full projects, create interactive visual experiences, and handle 3D tasks.
Launch Team
Atlassian LoomShow, don't just tell with Loom
Promoted
LongCat-2.5 is Meituan's latest step in that direction. It's a native multimodal model designed for long-horizon agentic work, combining visual understanding with stronger coding and engineering capabilities.
What's new:
Long-horizon coding: Trained on longer task trajectories to improve its ability to work across complex codebases and complete full-project builds.
Native multimodality: Processes visual and text inputs in a unified token-based architecture, enabling visual understanding alongside coding.
Visual development: Supports interactive web experiences, creative design, visual programming, and 3D tasks.
Agentic reinforcement learning: Uses large-scale reinforcement learning, synthetic training tasks, interactive environments, and verifiable tests to improve end-to-end task execution.
Multi-harness compatibility: Works with Claude Code, Codex, Hermes, and other agent harnesses.
Large-scale architecture: A mixture-of-experts model with 1.6 trillion total parameters and approximately 48 billion activated per token.
Who it's for: Developers, AI researchers, and teams building coding agents, autonomous engineering workflows, and multimodal applications.
Explore LongCat-2.5