Artificial Analysis Intelligence Index v5 makes 60% tests private
Artificial Analysis just rolled out its Intelligence Index v5, and it's shaking up the AI leaderboard.
With 60% of test questions now private to prevent models from "benchmaxxing," the practice of preparing models for evaluations by feeding them old questions available online in repositories like Github and (famously) Hugging Face,
Moonshot AI's Kimi K3.1 Pro slid from fourth to ninth place, a big shift for a model that was recently making headlines.
Google's Gemini 4 Carbon tops index
Google's Gemini 4 Carbon is now at the top, thanks to strong performance in complex reasoning tasks.
Meanwhile, Moonshot AI is holding off on releasing Kimi's model weights until November 4, as background context on open-weight model security risks.
NVIDIA CEO Jensen Huang weighed in, saying distillation is just competition, even as Anthropic worries about potential downsides.