Good and concrete suggestion. You’re right, right now the model selector has no tooltip at all, and that’s exactly where it makes sense to show that in the current deployment Kimi K3 doesn’t take images natively and instead works via a text description. That’s the kind of transparency this thread was missing.
I passed this to the team along with the rest of the feedback from this thread as one tracked issue, both adding a UI note tooltip or badge in the picker and making sure the model doesn’t present guesses as real observations.