Meituan open sources LongCat-2.0, the 1.6T, near-frontier agentic coding model that's been leading OpenRouter — trained entirely on Chinese chips

2026-06-30 · Source: VentureBeat · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Expert, medium

Summary

Meituan has open-sourced LongCat-2.0, a 1.6-trillion-parameter Mixture-of-Experts (MoE) agentic coding model, under an MIT license. This model, previously known as "Owl Alpha" and a top performer on OpenRouter, features a native 1-million-token context window. Crucially, LongCat-2.0 was trained entirely on a cluster of over 50,000 domestic Chinese ASICs, demonstrating the capability to scale near-frontier AI without reliance on U.S. Nvidia GPUs. Its commercial framework includes aggressive pricing, with a limited-time promotional rate of \$0.30 per million input tokens and \$1.20 for output, and uniquely offers free processing for context-cache hits. The model excels in software engineering tasks, scoring 59.5 on SWE-bench Pro, surpassing GPT-5.5's 58.6, and utilizes a Multi-Teacher Optimization via Mixture of Specialized Experts (MOPD) post-training framework for specialized agentic, reasoning, and interaction capabilities.

Key takeaway

For AI Engineers and Directors of AI/ML evaluating large language models for autonomous software engineering, LongCat-2.0 presents a compelling, cost-effective alternative. Its 1-million-token context window and zero-cost cache hits significantly reduce operational expenses for iterative coding tasks. You should consider integrating this MIT-licensed model to bypass data privacy concerns and high API costs of proprietary systems, especially for codebase migrations or continuous infrastructure optimization.

Key insights

Meituan's LongCat-2.0 demonstrates near-frontier AI can be developed on domestic Chinese ASICs with competitive performance and cost.

Principles

• MoE sparsity scales parameters while limiting active computation.
• Context-cache hits can be processed free of charge.
• MIT license offers maximum enterprise integration flexibility.

Method

LongCat-2.0 uses LongCat Sparse Attention (LSA) with Streaming-aware, Cross-Layer, and Hierarchical Indexing for 1M-token context. Post-training employs MOPD, segregating Agent, Reasoning, and Interaction Experts.

In practice

• Use zero-cost context caching for iterative agentic development.
• Deploy MIT-licensed models for deep integration into proprietary software.
• Leverage specialized expert clusters for compliance-sensitive tasks.

Topics

• LongCat-2.0
• Agentic AI
• Mixture-of-Experts
• Chinese ASICs
• Open-Source LLM
• Software Engineering
• Cost Optimization

Code references

• meituan-longcat/LongCat-2.0

Best for: CTO, VP of Engineering/Data, Entrepreneur, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Meituan Releases New AI Model Trained Entirely on Chinese Chips — Meituan's LongCat-2.0, trained on Chinese chips, signals growing domestic AI hardware independence.
• Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding — Ornith-1.0 demonstrates strong agentic coding capabilities using self-scaffolding LLMs built on open-source foundations.
• Zhipu AI's GLM-5.2 closes in on closed-source leaders in coding marathons — Zhipu AI's GLM-5.2 sets a new open-source benchmark for long-horizon coding with a 1-million-token context and novel architectural efficiencies.
• [AINews] Tasteful Tokenmaxxing — The AI landscape is rapidly advancing with specialized hardware, open-source models, and sophisticated agent platforms.
• DeepSeek v4 Agentic Coding with OpenCode | Building UI for Agentic RAG Template | 🔴 Live — DeepSeek V4 models offer competitive pricing via OpenRouter, but exhibit persistent "reasoning context errors" during agentic coding tasks.
• TAI #196: Quiet but Significant Agent Upgrades to Codex (Subagents) and Claude (Context) — Parallel agentic workflows and robust long-context processing are key to advancing AI development and deployment.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by VentureBeat.