News
How powerful is the official version of DeepSeek V4: The AI killing line is well-deserved and its price-performance ratio kills most of its competitors.
2 min read
Source: news.mydrivers.com
Kuai Technology reported on August 1 that on the last day of July, DeepSeek finally released the official version of V4. The first version is V4 Flash-0731. The official version of Pro will not be released until early August, along with the Harness tool of DeepSeek. After the release of V4 Flash-0731, netizens who had previously called Liang Wenfeng Xiaoliang and Liang Baikai because of the delay in DeepSeek immediately restored the title of Liang Sheng because its performance exceeded expectations. The official test performance exceeded GLM-5.2 and was lower than Opus 4.8. Some tests can even catch up with the latter. As for how powerful V4 Flash-0731 is, many netizens on the Internet have spontaneously tested it. It basically conforms to the official statement. The performance is amazing compared to its 284B parameters, and is stronger than many 1000B trillion models. If you still don’t understand the performance level of V4 Flash-0731, some netizens used GPT analysis to summarize the data on the AI performance ranking Artificial Analysis, and made a detailed comparison based on API pricing, which almost covers common large models at home and abroad. This list is very simple. The abscissa is the price, the lower the better, the ordinate is the performance score, the higher the better. Each large model is also distinguished according to the thinking gear. If you don’t want to compare one by one, you can just look at the position of V4 Flash-0731. The large model located in its lower right corner means that either the performance is not as good as V4 Flash-0731, or the price is not as good as it. To put it simply, the large domestic or foreign models in the lower right corner have been killed by V4 Flash-0731. There are mainly two types of models that have been lucky enough not to be killed by V4 Flash-0731. One is the one with a lower price. The main ones are GPT-5.6 Luna (low) and Xiaomi's MiMo-V2.5. The former has just been reduced by 80% in price, otherwise it will fall into the kill line. Xiaomi's large model has been reduced in price with the V4 preview version before, and it is also multi-modal, which has some advantages