If a @convex optimized fine-tuned model as smart as kimi, but 2.5x as fast and 30x cheaper would be interesting to you, lmk. Maybe these will start appearing in the Convex AI gateway soon.