LongCat-2.0

1.6T MoE trained entirely on AI ASICs

Open Source

Artificial Intelligence

Visit WebsiteSee on Product HuntHugging Face ⧉Twitter ⧉

The stable run on ASIC superpods with no irrecoverable spike is the genuinely impressive part here, more than the parameter count. For agent use the metric I care about is long-horizon adherence rather than single-turn tool accuracy: in our loops the model at tool call 30 has usually forgotten a constraint it agreed to at call 5, and 1M context helps recall without stopping that instruction decay. Did the agentic post-training target staying on a plan across many tool calls, or mostly one-shot tool-call correctness?

How does the 560B MoE setup actually perform on smaller hardware for fine tuning, or is it really only practical to run through Meituan's own infrastructure?

how does the 560B MoE setup hold up on longer context tasks compared to dense models like DeepSeek?

Curious how the 560B MoE setup actually feels in latency for real time stuff like chat compared to something like DeepSeek, and what kind of hardware you'd realistically need to run a distilled version locally?

Runs surprisingly snappy for a 560B MoE — I had low expectations but the reasoning responses came back fast and actually thought through edge cases instead of hand-waving.

the reasoning speed on long context prompts genuinely surprised me, felt closer to a smaller model than a 560B MoE.

Curious how this stacks up against other open-source MoE models on benchmarks like HumanEval or MATH, and is it actually free to deploy commercially or are there usage limits built in?

how does the 560B MoE setup handle latency on longer reasoning chains, and is there a hosted endpoint or is it self-host only?

How does the 560B MoE setup compare on inference cost versus a dense model of similar capability, and is there any self-hosting guidance for teams without massive GPU budgets?

Ran a coding question through LongCat-Flash-Thinking and the response came back fast with clean, working code instead of long-winded explanations. Really impressed by how snappy it feels for such a big model.

Tried it on a logic puzzle and the reasoning came back structured and fast, way smoother than I expected for a 560B model.

LongCat handled a tricky multi-step coding prompt way faster than I expected, and the reasoning chain felt genuinely clear instead of just confident. The MoE setup seems to pay off in practice.

Curious how this stacks up against DeepSeek or Qwen on coding benchmarks, and is the 560B MoE actually deployable on a single node or do you need serious infra to run it locally?

finally a chinese-built model that actually feels fast on long reasoning chains. really impressed how it handled a math-heavy prompt without losing the thread halfway through.

Pulled LongCat-Flash-Thinking for some debugging yesterday and the response time genuinely caught me off guard. Reasoning feels sharp without the usual lag.

Finally got around to testing LongCat-Flash-Thinking on some multi-step coding problems and the speed honestly caught me off guard, especially for a 560B MoE. The reasoning chains feel surprisingly tight too.

Ran it through some tricky math problems and the responses came back noticeably quick for a model this size. Curious to see how the open weights hold up on longer agentic workflows.

How does the 560B MoE setup actually translate to faster inference in practice compared to dense models of similar size, and is it open weights under a permissive license?

How does LongCat-Flash-Thinking handle context length on really long documents, and is it open weights or just open source code?