The model demonstrates strong capabilities given its relatively small size.
However, I have concerns regarding multilingual training, particularly with smaller models that may inappropriately incorporate cultural elements from specific countries associated with certain languages, sometimes causing them to shift away from the original prompt language.
This was observed during testing of the advertised model with the recommended settings, and it seems inconsistent for a system designed as "agentic" to possess such extensive vocabulary beyond English.
(my test was mentioning "kaomojis" as a word, which made the model speak Japanese, you can get what I am saying not in just this model but many others across other examples but that's how I tested the issue in this model )
While acknowledging occasional limitations in tool handling, this model remains competitive within its class, I personally think that we got another very promising lab contributing to opensource edge AI.