It's sad that they only compare to open weight models. I feel most users don't care much about OSS/not OSS. The value proposition is the quality of the generation for some use case.
I guess it says a bit about the state of European AI
It seems to be a reasonable comparison since that is the primary/differentiating characteristic of the model. It’s really common to also and seemingly only ever see the comparison of closed weight/proprietary models in a way that seems to act as if all of the non-American and open weight models don’t even exist.
I also think most people do not consider open weights as OSS.
It’s not for users but for businesses. There is demand for inhouse use with data privacy. Regular users can’t even run large model due to lack of compute.