Is Kimi K3 Really That Good?! (Don't Just Believe The Hype)

YouTube · Watch

• Custom-benchmarks Kimi K3 against Opus 4.8 on complex engineering tasks rather than relying on published benchmarks.
• Finds an 8% vs. 36% failure-rate gap between the models — a large real-world reliability difference publicly reported benchmark scores didn’t surface.
• Core argument: “public benchmarks can’t be trusted” for real-world reliability assessment — same skepticism-of-official-benchmarks thread this project tracks elsewhere.