Claude Opus 5.5 Max Tops Agent Leaderboard in Its First Appearance

The Arena platform has released its latest AI large model leaderboard. Rankings are based on anonymous blind test votes and ELO scores, reflecting user preference rather than absolute capability.

The big highlight this week: Anthropic's Claude Opus 5.5 Max topped the agent leaderboard in its first appearance, with a 15.53% net improvement rate, a 17.29% task confirmation success rate, and only a 0.37% tool hallucination rate. On the overall leaderboard, Google's Gemini 4 Argon High stays at No. 1 with 1525 points, while Anthropic holds the top four spots.

Chinese models sit in the upper-middle tier: Moonshot's Kimi K3 Max ranks 16th overall, and takes 10th on the coding leaderboard with Xiaomi's MiMo right behind at 11th. On the frontend leaderboard, two Alibaba Qwen models made the top 15, and Xiaomi's MiMo entered the list for the first time.

The results show Anthropic leads in agent and frontend capabilities, while Chinese models keep a steady presence in the upper-middle tier.