🚀 Introducing LongCat-Flash-Thinking-2601 — A version built for deep and general agentic thinking. ✨ Highlights: 🤖 Top Tier Agent Capabilities 🔹 Performance: Top tier benchmark results (TIR / Agentic Search / Agentic Tool Use) ; superb generalization ability, outperforming Claude in complex, random tasks 🔹 Env Scaling: Multiple automaticly... constructed high-quality environments; dense dependency graph 🔹 Multi-Env RL: Extended DORA (our RL infra), supporting large-scale multi-environment agentic training 🛡️ Real-World Robustness 🔹 Performance: Solid performance in messy, uncertain scenarios (Vita-Noise & Tau^2-Noise) 🔹 Noise Analysis: Systematically analyzed real-world noise in agentic scenarios 🔹 Curriculum RL: Increasing noise type & intensity while training 🎯 Heavy Thinking Mode ​🔹 Parallel Thinking: Expands breadth via multiple independent reasoning tracks 🔹 Iterative Summarization: Enhances depth by using a summary model to synthesize outputs, supporting iterative reasoning loops 📅 One more thing: 1M-token context via Zigzag Attention is coming soon. 🔍 Try it now: ✅ API access for this version is also available. Hugging Face: GitHub:show more

Congrats! But note that the and Kimi logos are the older versions.

Appreciate the reminder! 🙏 We’ll update to the latest logo in our technical reports and related media materials. Stay tuned for more updates, and we’re always open to further discussions ❤️

amazing models as always, is "Zigzag Attention is coming soon." the sparse attention here most likely yes but if so i'm very hype

You got it! Yes, that's exactly the paper. Stay tuned, we're hyped too!

congrats!!! the 8 thinkers in the UI is so fun :D

Thanks! So glad you like the 8 thinkers 😊

兄弟加油啊 连夜发 模型 等着你们明天干翻阿里呢 他们要拿你们的饭碗了！The stock price should go up now.

I think it's worth getting benchmarked @ArtificialAnlys :)

@ArtificialAnlys Agreed!

Great work!

congrats!! Good job @Meituan_LongCat

Parallel thinking?

omg thats crazy good

Thanks! More awesome on the way. 😄

You guys rock it! 👍

congrats!

Thanks!

Did Meituan actually deliver this time or what. The benchmarks point to just the kind of model I wanted. Will play around with it.

@fierceeyes3 Glad it matches what you're looking for!

Awesome stuff!

Congrats guys, you cooked 🔥

cool zigzag attention seems strong

@zvdom começou mesmo

Report when ?

The technical report will be released in the next few days.

That's wonderful! May I ask when the global standalone app will be released?

I wanted to see docs and pricing and I was asked for my phone number when I clicked "API Platform" Did you make the UX impossible on purpose?

🤩🤩🤩

Just whipped up a quick PPT to summarize the LongCat release – makes the key points way easier to digest!

Where you guys studying engineering

@huggingface I wanted to try the app but the app langauge is in Chinese .😅

@ArtificialAnlys We need normal+heavy on Artificial Analysis Index v4.0!

@Meituan_LongCat, this agentic focus in LongCat-Flash-Thinking-2601 should improve inference capabilities significantly.

longcat的网页端的深度思考已经更新到这个最新模型了吗？

@CWong24871 是的，已经更新到最新模型

This is moving agentic AI from benchmark wins to real-world reliability. Multi-env RL, noise-aware training, and long-context reasoning are exactly what scalable agents need.

Keep cooking china❤️

Thanks a lot! More to come.🚀

@Meituan_Tech LongCat-flash-agile 🙏, an updated version of longcat-flash-omni with multi-mode hybrid inference: base, thinking, parallel thinking which can be switched pre/post model loading. default mode: base LLM. PLEASE 🥺

It is pre-trained on over 50K AI ASIC superpods. Significant effort. Optimizations achieving over 35% training throughput. More interesting than the benchmarks.

真的太难用了，难道这个没和langchain适配么（狗头），之前用mino/minimax跑好好好的agent，切换了longcat根本执行不下去，看日志感觉json的指令遵循实在是太差了

Any way to use it inside copilot or well integrated into vscode? Opencode perhaps?

congrats! its a good model