AI Model Blind Test Rankings Update: Chinese Models Shine in Specific Categories

Arena released its AI model blind test rankings for week 36 of 2026, with several new models entering the list. In the overall chart, Anthropic's claude-fable-5 topped with 1507 points, followed closely by claude-opus-4-6-high. Newcomer claude-fable-5.1-max jumped straight to third place, while Google's gemini-3.8-flash-high ranked 8th. The top 10 are tightly clustered, with less than 30 points separating them.

Chinese models impressed in specific categories. In front-end development, Alibaba's qwen3.8-max-0902 debuted at 4th place, with Tencent's hy4-preview and Zhipu's glm-5.3-flash also entering the top 15. In AI video, Alibaba's wan3.0 debuted in the top three. In coding, Moonshot's kimi-k3-max ranked 6th, and Zhipu's glm-5.3-flash rose to 9th. Overall, overseas models still lead, but Chinese models are advancing quickly in niche areas.