How Far is DeepSeek V4 from the World's Strongest AI? American Researchers Release Results: 7 Months Behind
DeepSeekV4 has been released for a week, with three price reductions or discounts, and a visual large model has also been released, basically completing this generation's large model layout. Compared to last year's DeepSeekR1, the release of DeepSeekV4 has sparked more controversy, mainly because it lacks the previous R1's feeling of technological leadership. It also lags behind AI large models developed by American companies on various AI leaderboards, mainly the top models of Google, OpenAI, and Anthropic, such as GPT-5.5, Opus4.7, and Gemini3.1Pro.
How far is DeepSeek V4 from the strongest AI? This is also a matter of great concern to Americans. Recently, the Council on Foreign Relations organized 3 senior researchers to conduct a research report, and their conclusion is that V4 aligns with the U.S. assessment of being 7 months ahead of China.
DeepSeek V4 has not achieved the current most cutting-edge large models, and DeepSeek itself is aware of this. They mentioned in their paper that they are about 3-6 months behind the most advanced frontier models, which is roughly consistent with the American assessment of being 7 months behind.
DeepSeek previously pointed out in its research reports the problems facing domestic AI large models: on the one hand, the gap in computing power, and on the other hand, the gap in world knowledge. Ultimately, it requires burning money to purchase AI computing resources, and domestic companies also face various blockades and sanctions.
This leads to an unequal foundation on both sides. DeepSeek V4's Pro large model has only 1.6 trillion parameters, while Opus 4.6 and GPT 5.5 are considered to have 5 trillion or even 10 trillion parameters. The scale of the two sides is incomparable, and it is normal for performance to be inferior.
The good news is that domestic companies are still making rapid progress in both computing power and world knowledge. With time, the gap in hard power can be caught up, at least the gap will not be too large.
Whether the gap is 3-6 months or 7 months, it actually does not determine the success or failure of DeepSeek V4. Domestic AI has a significant advantage in open source and pricing, with costs only a few tenths or even hundredths of the top American large models. Ordinary people can also use it with confidence, and this is the key.