I’m generally fine with the models available in Kagi’s AI features, and the on-page tips as they are today already do a good job of educating. That said, for both explanatory and transparency purposes, I hope the documentation could elaborate on the reasons behind the selection.
For example, the models’ empirical or statistical advantages in performing specific tasks, the balance between cost and quality, the developing experiences with the providers, among others.
Indeed, I’m specifically interested in why the assistant mode uses Claude models exclusively. Other than being slightly lower priced and less whimsical, I found them overly terse and conservative, and the Claude 2 responses significantly slower than GPT-4 Turbo, making each search with the expert mode a painful wait.
