When I use the Kimi K2.5 model in Kagi Assistant, if the response starts with the word I, it always cuts off the word I.

Any response from the Kimi K2.5 model that starts with the word I cuts off the first word (so the first word I see is the word that follows I).

I've reproduced this on iPhone in Safari and on MacOS in Safari and Vivaldi.

See attached screenshot.

Any response that starts with the word I should not have the first word cut off.

Also happens in other languages. It's the first character of the response.

I get some interesting responses from this sort of prompt

This occurs on GPT OSS 120B too.

Clearly it meant to say "Apple's"