I gave Kimi K3 a shot at auditing my post-quantum crypto project, it found 5 real bugs Fable/Opus 4.8 and GPT-5.6 Sol had all missed

(Slides because why not) I've been building a serverless post-quantum group-encryption protocol for the last few weeks, mostly with Opus 4.8 and Fable, with GPT-5.6 Sol running adversarial review on the most challenging part (a recovery/finality mechanism). It went through four rounds of review: each round found something, Claude fixed it, and by the end the formal model was solid. I was trusting the process... But then Kimi K3 comes along and I figured I'd give it a shot. No guardrails on a cryptography project? Yes please. So I wired it up through OpenRouter and pi, and pointed it at the exact same model as an independent auditor. It found five real bugs the others had missed. It ran its own sweep and confirmed the core was sound, but I reproduced every one myself to make sure. Kimi K3 caught blind spots the other models couldn't see (or weren't allowed to). Very impressive!