Comment by curioussquirrel
Comment by curioussquirrel 5 hours ago
This is probably not a core concern for most HN readers, but at work we do multilingual testing for synthetic text data generation and natural language processing. Emphasis on multilingual. Gemini has made some serious leaps from 1.5 to 2.5 and now 3.0, and is actually proficient in languages that other models can only dream of. On the other hand, GPT-5 has a really mixed performance in a lot of categories.
Very good to know. I use Gemini for many translation related work, the 1m windows is very helpful too.