Google's Gemini 4 Carbon Tops LLM Leaderboard
Google's newly released Gemini 4 Carbon (High) debuted and claimed the top spot on the Arena LLM blind-test leaderboard for the 46th week of 2026 with a score of 1526. Meanwhile, Chinese models such as Moonshot's Kimi series and Alibaba's Qwen maintained stable positions in the top tier for specialized tasks like video generation and frontend development. This weekly ranking highlights the rapid iteration cycles of major AI labs and the shifting dominance among top-tier models like Google's Gemini, Anthropic's Claude, and emerging Chinese competitors. Tracking these crowd-sourced preference leaderboards helps developers and enterprises gauge real-world subjective performance across diverse workloads. Gemini 4 Carbon (High) scored 1526 points, narrowly beating its sibling model Gemini 4 Argon (High) at 1522 points within the margin of error. Anthropic maintained a heavy family footprint in the top 8 positions despite slight score declines for older iterations like Claude Opus 4.8.
## BACKGROUND
The Arena platform relies on crowdsourced blind pairwise comparisons, where users input prompts and vote on the better of two anonymous model responses. These votes are then processed using the Elo rating system to generate a competitive leaderboard reflecting current human preferences.