B站 has unveiled a major model competition - the "AI Infinite Arena" review ranking, and also released the first round ranking. In the real-device tests conducted by 10 UPs, GPT-6Astra took the top spot and won the "most climbs to the top" title, leaving GLM-5.3 behind; three of the top five models are domestic large models, competing closely with overseas giants.
This arena differs from traditional benchmarking lists. It gathers real-world tests of over a hundred large models by B站's various field UPs into a public square, covering topics such as coding, reasoning, collaboration, and knowledge, without fixed evaluation dimensions - creators from different sections independently propose topics based on real workflows, professional applications, or fun ideas, allowing models to demonstrate their actual differences in the same challenges. Mainstream models such as DeepSeek, Kimi, ChatGPT, Claude, Gemini, DouBao, Qianwen, Hy, and MiniMax are all included in the comparison, with rankings updating in real time and continuously opening up for registration across the entire platform.
