I was using this app entirely this day and to my surprise, it was pretty good. The model is much capable than I expected and the benchmarks also support this. The only con is that the context limit is 131k, if it has been 230k near to gemma 4 31b, it would have been unbeatable for the size. Anyway, looking forward to try more,
I've been using M.1 this evening through Openrouter with Hermes and its been so good for general tasks... fixing scheduled tasks, youtube analysis/summaries, and some script building, have all been very quick and clean. Very impressed so far!
It seems very efficient... in that it seems to get onto the task quickly, precise tool calls, doesn't fluff around when looking for files, that kind of thing.
Trained from scratch and (claiming to be) on par with SOTA open weight models!? I've never heard of this company at all - it seems easier than it should be to be model maker...
abhijithwarrier@reddit
I was using this app entirely this day and to my surprise, it was pretty good. The model is much capable than I expected and the benchmarks also support this. The only con is that the context limit is 131k, if it has been 230k near to gemma 4 31b, it would have been unbeatable for the size. Anyway, looking forward to try more,
Tasty_Tackle_6059@reddit
I've been using M.1 this evening through Openrouter with Hermes and its been so good for general tasks... fixing scheduled tasks, youtube analysis/summaries, and some script building, have all been very quick and clean. Very impressed so far!
It seems very efficient... in that it seems to get onto the task quickly, precise tool calls, doesn't fluff around when looking for files, that kind of thing.
Very interested to see the pricing.
MokoshHydro@reddit
Was playing with in kilocode yesterday. They look really good (competive with Qwen), the only downside is small context size.
ComplexType568@reddit
waiting for the gguf
rm-rf-rm@reddit
Trained from scratch and (claiming to be) on par with SOTA open weight models!? I've never heard of this company at all - it seems easier than it should be to be model maker...
coder543@reddit
Always cool to see more open weight models!
pmttyji@reddit
Good to see one more 30B range MOE model. That seems honest benchmark(Check the graph)
abkibaarnsit@reddit (OP)
XS.2 is currently out. 33BA3B (https://huggingface.co/poolside/Laguna-XS.2)
M.1 still not open. 225BA23B
Free on OpenRouter M.1 , XS.2