Bilibili Launches AI Arena Model Ranking, GPT-6 Astra Tops First Round

Chinese video platform Bilibili has launched the 'AI Infinite Arena' large model ranking and released its first round of results. GPT-6 Astra took the top spot across 10 creator reviews, beating GLM-5.3 for the most first-place wins. Three Chinese-made models ranked in the top five.

The arena gathers real-world tests by Bilibili creators on over a hundred large models, covering coding, reasoning, collaboration, and knowledge topics. Unlike traditional benchmarks, it sets no theme or dimension limits. Creators design their own tasks based on real workflows, professional use cases, or fun ideas, with head-to-head comparisons showing actual performance differences.

The ranking covers mainstream models including DeepSeek, Kimi, ChatGPT, Claude, Gemini, Doubao, Qwen, Hy, and MiniMax. Rankings update in real time, and sign-ups remain open. Over the past year, AI knowledge content watch time on Bilibili grew 72% year-on-year, with over 190 million monthly viewers of AI content.