Artificial Analysis version 5 of the Intelligence Index reshuffles the AI leaderboard, and Google's Gemini 4 Carbon now holds the top spot. The update introduces private datasets and agentic task evaluations, which closes the loophole where labs trained models directly against public benchmark data. The gap between US frontier models and open weight alternatives widened under these conditions.
The methodology change is the real story here. Private datasets mean benchmark contamination is harder, and agentic tasks test multi-step reasoning under realistic conditions rather than single-prompt recall. Those two changes alone explain why the rankings moved, and why previous leaderboard positions should now be read with skepticism.
Google and Microsoft cloud earnings provide a revenue backdrop for the model arms race, and a DOJ probe into NVIDIA adds a regulatory dimension worth tracking. Read the full breakdown for NLW's analysis of what the widening US-versus-open-weight gap means for the next wave of model releases.
[WATCH ON YOUTUBE →]