Moonshot dev explains why Kimi K2 went open-source — and why ads stopped

teortaxesTex · x · 2026-10-24

A long essay apparently by a Moonshot AI developer explains the deep logic behind open-sourcing Kimi K2 and reveals several inside details.

Why open-source: reputational gain (K2 wouldn't trend if closed); community leverage (MLX ports, 4-bit quantization within 24h); and crucially, openness forces higher technical standards — you can't hide hacks when third parties must reproduce your numbers with the same weights, whereas closed services can stack dozens of models and countless workflows and call it MoE.
Ad controversy: after pausing paid ads early this year, searching "kimi" in Chinese app stores didn't even show Kimi on page one; Apple's App Store suggested "Doubao" instead.
On DeepSeek-R1: colleagues saw it as proof that raw capability is the best marketing; their only regret was not proving it first.
Strategy: in January, CEO Yang Zhilin went further than the author's radical proposals — kill K1 updates, put all resources into fundamentals and K2. On the "pivot to Agents" chorus, the author argues most Agent products are nothing without Claude, and intelligence is still capped by the model in 2025.

The essay also criticizes Kai-Fu Lee's "I care about AI ROI" remark, framing AGI pursuit as a single-log bridge where hesitation is fatal.

More from Companies & People