“In a debugging task, KIMI K3 could not identify or fix a bug, while Fable-5 and GPT-5.6 both solved it in one attempt.”