Mistral Small but Mighty - Apache 2.0, Multimodal & Fast

Looks great. Can you include a bit of multi-language tests? Maybe use google translate on outputs, that would be awesome

thanks for the great content and please keep up the good work

Would have liked to see more info about the hardware needed to run these models at home - and your thoughts on models/hardware which can be run at home - but still a great video - cheers!

It's a promising model and kudos to Mistral for the Apache license. I just wish they'd also trained smaller edge versions like Google has done with Gemma 3. I mean, the minimum GPU you need for his is a 3090. And even then you won't be able to fully utilize the 128k context window. That makes me wonder if Mistral is intentionally pushing people toward using its API for synthetic data gathering.

I would have loved a 12b version, my 3080 10gb can't run a 24b model...

im really interested in llm which fit in 24GB of vram. i think a comparison between llama3, phi4, gemma3 and this would be great!

great video man
because of your videos I finished one production standard project in our company
the one you explained context caching helped me a lot

Looking forward to the comparison between Mistral Small v3.1 and Gemma 3 27B