LongCat-2.0 - 1.6T MoE trained entirely on AI ASICs
LongCat-2.0 is an MIT-licensed 1.6T-parameter MoE model with ~48B active parameters, 1M context, LongCat Sparse Attention, and post-training for coding and agentic workflows. It was trained on AI ASIC superpods and integrates with Claude Code, OpenClaw, and Hermes.
Replies
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?
the reasoning speed on long context prompts genuinely surprised me, felt closer to a smaller model than a 560B MoE.
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.
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?
how does the 560B MoE setup hold up on longer context tasks compared to dense models like DeepSeek?
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?
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 LongCat decide which parts of a conversation are most important to retain over time?
Huge congrats on the launch! Testing models locally is a huge part of my workflow right now, so having another heavy-reasoning, open-weight option is fantastic. I'm curious—how does the prompt caching perform in terms of Time To First Token when maxing out the context window on standard hardware, like a 48GB Mac? Excited to benchmark this!
Congrats on the launch. I gave LongCat a quick try and noticed one small thing that might be worth checking.
I had the language set to English, but one of the responses came back in a non-English language. It may be an edge case, but it could confuse first-time users who expect the selected language to carry through the chat.
Hope this helps.