I asked a Finnish person how good an answer about the language example from ChatGPT was. It turned out to be a hallucination, a confidently sounding nonsense.

The quality of internet trained models degrade very fast with language material size

Hallucinations and output quality are two different problems. Hallucinations are usually expressed in perfect sounding sentences by the LLM, that's what makes it so convincing to end users.

I live in Finland, ChatGPT is great at Finnish and far better at it than EuroLLM.

EuroLLM doesn't even beat the open source Gemma 2, and Gemma 3 is already out.

Maybe things have improved since that time (about a year ago). But my question was along the lines of "is it true that in Finnish this word means that?" and ChatGPT promptly confirmed with nonsense

>All the models from all the big providers (even the Chinese models!) support all of these languages already.

As a speaker of one of those languages...no, they definitely don't. You can get some words out, but it's not necessarily anything sensible