We're benchmarking a bunch of models for one of our internal agents. Kimi K2 is up to 5x faster and 50% more accurate than frontier proprietary modelsā¦ And it took very little work to switch out the models via @aisdk / @vercel ai gateway (served by @groqinc)
