DeepSeek Big update tonight. Is the house really collapsed?

On the other hand, many people praised DeepSeek’s new model V4 Pro. The official version is really awesome. It is the release of Fable 5 level. The nuclear bomb was so scared that it collapsed on Ultraman. On the other side, netizens complained about the price increase, the model capabilities seemed to be lower than expected, and they were even suspected of deploying the wrong model. . . Although this update of the official version of DeepSeek V4 Pro is not the most explosive, it is definitely the one with the most divided reviews. Is it Liang Sheng or Liang Baikai? This wave seems to be Schrödinger's beam. . . But it can only be said that in the past period of time, what everyone has been able to experience is actually an incomplete version of DeepSeek. Careful netizens must have noticed that starting from the official version of V4 Flash, below the official model score table of DeepSeek, there will be a line of small words telling everyone: These results were run in DeepSeek Harness, or DSH for short. Here, I would also like to give some friends a brief introduction to Harness. Let's put it this way, the big model itself is just an extremely smart brain in a vat. If you want this smart brain to do work, such as writing a piece of code for you, checking information, or using various tools to complete complex tasks, you must equip it with a mecha and a nervous system. In the AI ​​industry, this exoskeleton mecha that connects tools to a model and manages its running status can be called Harness. The same brain, that is, the model, actually displays different capabilities under different harnesses such as Codex, ClaudeCode, Workbuddy, etc. Therefore, the entire AI circle is also full of interest in this DeepSeek Harness. Why can it improve the model running score so much? And just now, the full-blooded DeepSeek puzzle was finally put together, and the Harness tool that had been waiting for so long was finally released. With it, DeepSeek