Bilibili launches AI model evaluation platform with gpt-6 leading the rankings

Bilibili has launched an AI model evaluation platform called 'AI Infinite Arena,' featuring a leaderboard for large language models (LLMs). The platform's first round of rankings showed GPT-6 Astra leading with a score of 10 out of 10, surpassing GLM-5.3. Three Chinese models made it into the top five. The platform aggregates evaluations from Bilibili creators, covering various real-world applications such as coding, reasoning, and collaboration. Unlike traditional benchmarking, the platform allows creators to set their own challenges, showcasing the practical performance of different models. The leaderboard includes models like DeepSeek, Kimi, ChatGPT, and Qwen, with real-time updates. Bilibili noted a 72% year-over-year increase in AI content consumption, with over 190 million monthly users watching AI-related content. The platform is open to all Bilibili creators, encouraging ongoing participation and updates. Bilibili highlighted the growing interest in AI, with a wide range of content from creation to user interaction. The platform aims to provide a comprehensive view of AI model capabilities through diverse and real-world testing scenarios.