British Reuters reported a few days ago that in the latest benchmark tests on various mainstream artificial intelligence models around the world, Chinese artificial intelligence companies are deeply searching for the lowest operating cost of the newly released DeepSEEK-V4-Flash model. The summary of the report is as follows: DeepSeek officially released the DeepSEEK-V4-Flash model on July 31, once again showing that it can provide an artificial intelligence alternative at an ultra-low cost. The DeepSeek-R1 model launched by the company attracted global attention when it was released in early 2025, raising questions in the market about the huge investment of US companies in the field of artificial intelligence. According to an evaluation by the American research institute Artificial Intelligence Analysis Network, the DeepSEEK-V4-Flash model charges 0.14 US dollars per million input terms and 0.28 US dollars per million output terms, and the average cost of completing each test is about 0.03 US dollars in terms of data units measuring artificial intelligence usage. Meanwhile, the GPT-5.6 Sol model of the US Open Artificial Intelligence Research Center costs 1.86 US dollars, and Anthropic's “Claude Fables 5” model is 3.15 dollars, which is 100 times the operating cost of the DeepSEEK-V4-Flash model. The comparison method adopted by the artificial intelligence analysis network not only takes into account model pricing, but also includes the amount of data required to process and generate to complete the task, so it can better reflect actual usage costs. The score of the DeepSEEK-v4-Flash model is comparable to Google's “Gemini 3.6 Lightning” model in the “Intelligence Index”, which combines programming, reasoning, and simulated workplace tasks, but lower than Kimi K3 of Beijing Moon Dark Side Technology Co., Ltd. and some models of OpenAI and Anthropic.

British Reuters reported a few days ago that in the latest benchmark tests on various mainstream artificial intelligence models around the world, Chinese artificial intelligence companies are deeply searching for the lowest operating cost of the newly released DeepSEEK-V4-Flash model. The summary of the report is as follows: DeepSeek officially released the DeepSEEK-V4-Flash model on July 31, once again showing that it can provide an artificial intelligence alternative at an ultra-low cost. The DeepSeek-R1 model launched by the company attracted global attention when it was released in early 2025, raising questions in the market about the huge investment of US companies in the field of artificial intelligence. According to an evaluation by the American research institute Artificial Intelligence Analysis Network, the DeepSEEK-V4-Flash model charges 0.14 US dollars per million input terms and 0.28 US dollars per million output terms, and the average cost of completing each test is about 0.03 US dollars in terms of data units measuring artificial intelligence usage. Meanwhile, the GPT-5.6 Sol model of the US Open Artificial Intelligence Research Center costs 1.86 US dollars, and Anthropic's “Claude Fables 5” model is 3.15 dollars, which is 100 times the operating cost of the DeepSEEK-V4-Flash model. The comparison method adopted by the artificial intelligence analysis network not only takes into account model pricing, but also includes the amount of data required to process and generate to complete the task, so it can better reflect actual usage costs. The score of the DeepSEEK-v4-Flash model is comparable to Google's “Gemini 3.6 Lightning” model in the “Intelligence Index”, which combines programming, reasoning, and simulated workplace tasks, but lower than Kimi K3 of Beijing Moon Dark Side Technology Co., Ltd. and some models of OpenAI and Anthropic.