LongCat-2.0
LongCat-2.0 is a next-generation, open-source MoE (Mixture of Experts) model with 1.6T total parameters, designed for agentic coding and featuring native 1M token context support. It achieves state-of-the-art performance on software engineering and terminal benchmarks, and is available on longcat.ai and OpenRouter.
Key facts
What is LongCat-2.0?
Industry-first trillion-parameter open-source model that delivers state-of-the-art agentic coding with dynamic compute and native 1M context.
Who is LongCat-2.0 best for?
Developers, AI researchers, software engineering teams
What are its main limitations?
- Requires 50,000-card domestic compute cluster for full training (not accessible to all)
- Model size may limit deployment on consumer hardware
Limitations are based on product materials or editorial synthesis; confirm on the official site.
Key features
- ✓1.6T total parameters with dynamic activation of 33B–56B per token
- ✓Native 1M token context with Linear-complexity Sparse Attention (LSA)
- ✓Zero-computation experts + ScMoE for token-level dynamic compute
- ✓MOPD (Multi-Teacher On-Policy Distill) multi-expert fusion
- ✓SWE-bench Pro 59.5 – leads Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.6
- ✓Terminal-Bench 70.8 – stable execution and error recovery in real terminal environments
- ✓Open-source on longcat.ai and OpenRouter – top 3 globally by call volume
Use cases
- →Agentic coding and software engineering
- →Complex terminal-based automation and error recovery
- →Long-context document understanding and code analysis
- →Multi-expert reasoning for challenging technical tasks
Pros
- +Industry-leading benchmark scores on SWE-bench Pro and Terminal-Bench
- +Efficient dynamic compute – simple tasks cost near zero compute
- +Fully open-source with active community support
Cons
- −Requires 50,000-card domestic compute cluster for full training (not accessible to all)
- −Model size may limit deployment on consumer hardware
Positioning
- Core value: Industry-first trillion-parameter open-source model that delivers state-of-the-art agentic coding with dynamic compute and native 1M context.
- Ideal for: Developers, AI researchers, software engineering teams
- Product type: Web app, API (OpenRouter)
Frequently asked questions
What is LongCat-2.0?
LongCat-2.0 is a 1.6T-parameter open-source MoE model designed for agentic coding, with native 1M context support and dynamic compute per token.
How does LongCat-2.0 achieve dynamic compute?
It uses Zero-computation experts + ScMoE: simple tokens cost nothing, while complex tokens get more resources, optimizing efficiency.
What benchmarks does LongCat-2.0 excel at?
It scores 59.5 on SWE-bench Pro and 70.8 on Terminal-Bench, outperforming models like Gemini 3.1 Pro and GPT-5.5.
Is LongCat-2.0 open-source?
Yes, it is open-source and available on longcat.ai and OpenRouter, ranking top 3 globally by call volume.
What is MOPD in LongCat-2.0?
MOPD stands for Multi-Teacher On-Policy Distill, a training technique that fuses Agent, Reasoning, and Interaction experts into a unified model.
Can I try LongCat-2.0 online?
Yes, it can be used online at longcat.ai and via OpenRouter.
You may also like
Curated suggestions based on similar categories.
FacialHarmonyAI
FacialHarmonyAI provides a private facial proportion study from one clear front-facing photo, highlighting symmetry, facial thirds, jaw and eye balance, and golden-ratio cues. Landmark detection runs in the browser and the photo is not uploaded to the cloud. A free preview is available, with an optional $9.99 one-time full atelier report.
RedditFind
AI Reddit growth agent that drafts native posts and researches communities for marketing and product insights.
Bonemeal
AI growth engineer for SEO and content automation
Zens AI
AI & developer tools - AI customer support for SaaS
StudyEdge AI
StudyEdge AI is an AI study coach that automatically builds your weekly schedule, plans every session minute-by-minute, and runs study sessions with flashcards, quizzes, and active recall. One app for the entire semester.
ShogunAI
Shogun AI is a set of generative AI and optimization features from Shogun, a no-code ecommerce page builder. It helps established merchants with complex needs scale store content more efficiently and was released alongside Shogun Insights as the first milestones in an expanded product vision.