Arena Weekly: Anthropic Models Dominate, Chinese Models Close In

Arena released its blind test rankings for week 33 of 2026. Anthropic's new claude-opus-4 models dominated the top five. claude-fable-5 stayed first with 1506 points. New entrants claude-opus-4-6-high and claude-opus-4-7-high took second and third. The top three are statistically tied. Google's gemini-3.7-flash-high debuted at ninth.

Chinese models are in the mid-to-upper range. Moonshot's kimi-k3-max rose to 12th, the best among Chinese models. Alibaba's qwen3.8-max is eighth, down two spots. In front-end development, kimi-k3-max holds second, just 18 points behind the leader. Alibaba's qwen3.8-max rose to third. DeepSeek's deepseek-v4-pro-high debuted at tenth. In coding, Anthropic models also lead. kimi-k3-max stays sixth, the top Chinese model. In AI image generation, Microsoft's mai-image-2.6-preview debuted at second. Chinese models seedream-5.0-pro and qwen-image-3.0-pro rank eighth and ninth. The rankings are based on anonymous votes, reflecting user preferences.