Wow, @ arena is made by Taiwanese people? The recent AI coding rankings are worth watching But I think the focus is no longer on 'who is first'. What really matters is that the top spot in AI models is becoming less scarce. ⚡ ️ Previously, people thought that AI would be a winner takes all: GPT-4 is far ahead, Other models can only catch up. But now when you look at rankings like Arena, the front row models are becoming increasingly crowded. Claude、OpenAI、Google、GLM、Qwen、Kimi， Various open source and closed source models are squeezed into the same capability range. The narrowing of the Elo score gap indicates that the model's ability is rapidly standardizing. This matter is very similar to hydroelectric power You turn on the faucet and don't really care which company supplies the water. What do you care about -Isn't it cheap? -Is it stable or not? -Will it break? -Can it be integrated into your workflow? AI models are also moving in this direction. When the gap in model capabilities narrows, what the market needs to reprice is not 'who is the smartest', but: Who can integrate into the workflow Who has the lowest inference cost Who can achieve corporate compliance and security Who has data feedback and user retention Who can turn a model into a product instead of a demo Especially the coding model is more obvious. Engineers may not necessarily choose 'first place on the list' in the end. He will choose the tool that is the most stable, cheapest, understands his codebase the most, and is least likely to suddenly break down. That's also why when I look at Arena rankings, the first thing I see is not rankings, but structural changes. The more crowded the top twenty, the thinner the moat representing the model itself. Value will shift towards products, materials, distribution, computing power costs, and enterprise deployment capabilities. ️ There is another interesting thing: Wei Lin Chiang, one of the core founders of Arena, a global AI evaluation infrastructure, came from a background in computer science at National Taiwan University. Later, he conducted research at UC Berkeley and developed the Chatbot Arena, a human blind testing and evaluation system. The strongest narratives in AI in the past were usually chips, servers, and supply chains. But the Arena incident reminds us: Not only can it provide hardware infrastructure for AI. You can also participate in AI's trust infrastructure. The most important issue for the future AI industry may not be: Whose model is the strongest But rather: Who is qualified to define what is strong Who can become a credit rating agency for the model world Who can make the market believe that these AI rankings are real? '' Two years ago, the strongest model itself was a moat. But in the next stage, what may be truly valuable is: -How are models evaluated -How was it deployed -How to be trusted -How to be used by enterprises for a long time. The war of AI is transitioning from 'model capability' to 'infrastructure' stage.