Compared to the previous Mistral Large, much more effort was dedicated to alignment and instruction capabilities. On WildBench, ArenaHard, and MT Bench, it performs on par with the best models, while being significantly less verbose. (4/N)

→ View original post on X — @guillaumelample,