Kimi K3 tops Code Arena and reignites the debate over AI economics

Exponential View (Azeem Azhar) · rss · 2026-07-23

Azeem Azhar argues that Kimi K3 has changed the conversation around AI economics after becoming the first Chinese model to top the frontend Code Arena benchmark. The newsletter says Moonshot’s earlier Kimi K2.6 and Alibaba’s coming 2.4T-parameter open-weight Qwen3.8 show Chinese labs are doing more with less compute, with an estimated 4–7x efficiency advantage over U.S. labs.

The piece also pushes back on the claim that Kimi K3 breaks the economics of AI. Instead, it suggests frontier open models are still only 4–7 months behind in cyber capability, and that lower task costs do not kill the business case for AI. Microsoft engineers are reportedly testing Kimi K3 inside Copilot, highlighting how quickly strong open models can spread through the ecosystem.

Original post →

More from Venture

Venture channel →