Sources
Matthew Berman Runs the Adversarial Test on Kimi K3's Fable-Beating Claims
Berman published 'Did Kimi K3 really beat Fable?' on July 18 (12 minutes, ~72K views), directly interrogating the benchmark claims from Moonshot's July 16 release rather than restating the press cycle. The framing matters because K3's headline win — first on Arena's Frontend Code eval — is a single-category result sitting alongside third place on GDPval-AA v2, and Berman's channel history shows benchmark-vs-hands-on gaps are where the real signal lives for builders choosing a model.
↳ Follow the thread