--- title: "LongCat 2.0 Is Free in Kilo for a Limited Time" url: https://daily.dev/posts/longcat-2-0-is-free-in-kilo-for-a-limited-time-pr0yqscmo source_url: https://blog.kilo.ai/p/longcat-20-is-free-in-kilo-for-a type: article source: "Kilo Blog" published: 2026-08-21T19:33:56.110Z updated: 2026-08-22T19:00:34.320Z tags: ["llm", "ai-coding", "mixture-of-experts"] reading_time: 5 upvotes: 0 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # LongCat 2.0 Is Free in Kilo for a Limited Time **[Kilo Blog](https://daily.dev/sources/kilo-ai-blog)** · 5 min read · 0 upvotes · 0 comments ## Summary Meituan's LongCat 2.0, previously the mysterious stealth model known as Owl Alpha, is now available for free for a limited time on Kilo. The 1.6-trillion-parameter MoE model activates roughly 48B parameters per token, is MIT-licensed on Hugging Face, and uses sparse attention plus specialized expert routing to keep costs down. It reportedly outperforms a Claude Opus reference on IFEval and IMO-AnswerBench while trailing on SWE-bench Pro, and costs less per completed task than DeepSeek's V4 Pro flagship. Kilo frames this as part of its model-agnostic, model-choice philosophy alongside top performers like GPT-5.6 Sol and Grok 4.6. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What model was the mysterious stealth model called Owl Alpha on OpenRouter and Kilo leaderboards? Owl Alpha turned out to be LongCat 2.0 from Meituan, a food delivery company, not a major AI lab as many had guessed. It was revealed at the tail end of Q2 after climbing leaderboards anonymously and consuming over 10 trillion tokens during its stealth period, many from Kilo users. _Track which stealth models turn out to be worth adopting for agentic coding by following model news on daily.dev._ ### How many parameters does LongCat 2.0 from Meituan activate per token? LongCat 2.0 is a 1.6-trillion-parameter Mixture-of-Experts model that activates roughly 48 billion parameters per token, with Zero-Compute Experts keeping per-token activation between about 33B and 56B. It uses LSA sparse attention for long context without quadratic cost blowup and routes work across separate Agent, Reasoning, and Interaction expert groups. It is MIT-licensed on Hugging Face. _Compare open-weight model architectures like this one on daily.dev before picking one for agentic coding._ ### How does LongCat 2.0's cost per task compare to DeepSeek V4 Pro 0813? LongCat 2.0 costs $0.12 per Intelligence Index task on Artificial Analysis's weighted cost-per-task metric, compared to $0.25 per task for DeepSeek's V4 Pro 0813, more than double. Both are open-weight 1.6T MoE models with 1M-context ambitions and similar architectures on paper, making LongCat notably more efficient per completed task rather than simply cheaper. _Weigh cost-per-task tradeoffs like this on daily.dev when deciding which model powers your coding agent._ --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-coding](https://daily.dev/tags/ai-coding), [#mixture-of-experts](https://daily.dev/tags/mixture-of-experts) [View this post on daily.dev](https://daily.dev/posts/longcat-2-0-is-free-in-kilo-for-a-limited-time-pr0yqscmo)