LongCat
Frontier reasoning models from Meituan
159 followers
Frontier reasoning models from Meituan
159 followers
LongCat is the AI model series from Meituan, featuring powerful, efficient, and open-source models for complex tasks. Its latest release, LongCat-Flash-Thinking, is a 560B parameter MoE reasoning model that sets a new standard for speed and performance.
This is the 2nd launch from LongCat. View more
LongCat-2.0
Launching today
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.
Free
Launch Team
Flowtica Scribe
Hi everyone!
LongCat-2.0 is a 1.6T-parameter MoE model with about 48B active parameters per token, 1M context, and open weights under MIT.
But it was not trained in the usual Nvidia-heavy way. The full training run was built on AI ASIC superpods, over more than 35T tokens, with no rollback or irrecoverable loss spike.
Training a trillion-scale model is already hard. Getting that run stable on alternative hardware is probably the more interesting story here 🤔
the "no rollback, no irrecoverable loss spike" claim on ASIC superpods is the part that would make me trust this more than the raw parameter count. that's usually where alternative-hardware training runs quietly fall apart, not in the benchmarks people see later. one thing I'd want to know before betting an agent pipeline on this: since it integrates with Claude Code and OpenClaw, did the ASIC training path change the numeric format at all versus a typical bf16 GPU run, and if so does that show up as any weirdness once you quantize down for inference, or does it behave like any other MoE checkpoint at that point
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?
Congrats to the LongCat team!
As someone based in China, Meituan is a household name to me — I use it almost every week for food delivery and local services. So it’s honestly a bit surprising, in a good way, to see the same company open-source a 1.6T MoE model under MIT. That contrast makes this launch especially interesting.
I’m also exploring AI-assisted PM workflows and building my own agent skills, so the combination of open weights, 1M context, and post-training for coding/agentic workflows really stands out. Curious if the team has any best practices for using LongCat-2.0 in long-horizon agent workflows where the context includes product docs, code, and task history.
How does the 560B MoE actually handle latency in real-world agent loops, and is there a hosted endpoint or do we need to self-host the whole thing to use it?
Curious how the 560B MoE actually handles rate limits in practice, like does Meituan expose a hosted API or are we expected to self-host the weights for any serious throughput?
How does LongCat-Flash-Thinking handle context length on really long documents, and is it open weights or just open source code?