LongCat 2.0
LongCat 2.0 is Meituan's sparse mixture-of-experts model — 48B active of 1.6T total. Built for coding, repository-level changes, long-horizon problem solving, and agentic workflows, with a ~1M token context window.
Technical Specifications
Live data fetched from the OpenRouter model registry — always up to date.
What is LongCat 2.0?
LongCat 2.0 is a sparse mixture-of-experts language model from Meituan with 48B active parameters out of 1.6T total. The sparse MoE design means only 48B parameters fire per token — delivering near-frontier capability at a fraction of the compute cost.
It is purpose-built for demanding engineering workflows: repository-level changes, long-horizon problem solving, and agentic workflows. On OpenRouter it is already a top choice for coding agents — powering billions of tokens from tools like Hermes Agent, Kilo Code, and Halluna.
With a 1,048,756-token context window and up to 262,144 output tokens, LongCat 2.0 can hold an entire codebase, plan multi-file changes, and execute long tasks without losing the thread.
Features & Capabilities
1.6T MoE
Sparse mixture-of-experts — 48B active of 1.6T total parameters.
Repo-Level Coding
Handles repository-level changes and complex refactors.
1M Context
1,048,756 token context — entire codebases in one session.
Agentic Ready
Tool calling built in — designed for long-horizon agent workflows.
262K Output
Up to 262,144 completion tokens per single response.
Great Value
$0.30/$1.20 per M tokens — strong capability at a low price.
Who Is LongCat 2.0 For?
Agentic Coding
Long-horizon coding agents that plan, edit, and iterate across a codebase.
Repository Changes
Apply multi-file refactors and repository-level modifications.
Long Problem Solving
Complex problems that need sustained reasoning over long sessions.
Tool-Using Workflows
Function calling for agents that browse, search and act on results.
LongCat 2.0 Pricing
- Frontier-class coding at a low price
- 60% launch discount on listed rates
- On Hoodgen AI — covered by your plan credits
- Tool calling for agentic workflows
What Makes LongCat 2.0 Different?
LongCat 2.0's edge comes from its sparse architecture: 1.6T total parameters with only 48B active per token. This delivers the capability of a frontier-scale model at a fraction of the inference cost — which is why its OpenRouter pricing sits at just $0.30/$1.20 per million tokens.
The model is tuned for engineering reality — repository-level changes rather than single-file answers, long-horizon problem solving that keeps context over many steps, and native tool calling for agentic loops. That is why coding agents route billions of tokens through it.
Its 262K max output also matters: long tasks that need generated code, plans, and analysis in a single response don't get truncated mid-stream.
Trusted by Agent Builders
LongCat 2.0 powers serious agentic workloads on OpenRouter — from open-source agent frameworks to IDE coding assistants — with high uptime (100% 3-day uptime, 99.84% availability) and fast 23 tok/s throughput.
LongCat 2.0 Privacy & Data Policy
LongCat 2.0 is developed by Meituan and hosted on OpenRouter through AtlasCloud. Prompts and completions are processed by the hosting provider's API. As with any third-party model, avoid sharing sensitive personal data.
Hoodgen AI always keeps free alternative models available so you can switch without interruption.
LongCat 2.0 vs Other Models
Frequently Asked Questions
Ready to ship serious code with LongCat?
Chat with LongCat 2.0 on Hoodgen AI — sign in and start a conversation in seconds.
Start Chatting Free