DeepSeek V4 has been released for a week. During this period, the price has been reduced or discounted three times, and a large visual model has been released. The layout of this generation of large model is basically completed. Compared with last year's DeepSeek R1, this time DeepSeek V4 has been more controversial after its release. The main reason is that it does not have the feeling of technological leadership like the previous R1. It actually lags behind the large AI models developed by American companies on various AI lists, mainly top models from Google, OpenAI and Anthropic, such as GPT-5.5, Opus 4.7 and Gemini 3.1 Pro.
How far is DeepSeek V4 from the strongest AI? Americans are also very concerned about this matter. Recently, the US Council on Foreign Relations organized three senior researchers to make a research report. They concluded that V4 is in line with the judgment that the United States is 7 months ahead of China.
DeepSeek V4 has not achieved the current top large model. This is something DeepSeek itself has realized. They mentioned in the paper that it is about 3-6 months away from the most advanced cutting-edge model, which is about the same as the 7 months behind the American judgment.
DeepSeek has also pointed out in previous research reports the problems faced by large domestic AI models. On the one hand, there is a gap in computing power, and on the other hand, there is a gap in world knowledge. In the end, money will be spent to buy AI computing power resources. Domestic companies also face various blockades, sanctions and other difficulties.
This leads to the unequal foundation of the two sides. The Pro large model of DeepSeek V4 only has 1.6 trillion parameters, and Opus 4.6, GPT 5.5 is considered to be a parameter quantity of 5 trillion or even 10 trillion levels. The size of is incomparable, and it is normal for the performance to be inferior.
The good news is that regardless of computing power or world knowledge, domestic companies are still making rapid progress in these two aspects. If the lag in hard power takes time, they can catch up, at least the gap will not be too big.
Whether it is a 3-6 month gap or a 7-month gap, it actually does not determine the success of DeepSeek V4. Domestic AI has great advantages in open source and pricing. The cost is only a few tenths or even one percent of the top American models . Ordinary people can use it with confidence. This is the key.
