According to CoinMeta, DeepSeek scored a comprehensive score of 69 in tests conducted by independent evaluation institutions artificial and analysis, ranking 17th and coming close to the top in factual question answering capabilities. This evaluation included 27 search products, with exa and perplexity taking the lead with both scoring 80 points. DeepSeek charges only $0.75 for every 1000 successful searches, making it the cheapest among all the paid search products API that participated in the test. After integrating with DeepSeek, its comprehensive score increased from 33 (when not connected to the internet) to 69. It performs well in directly finding facts, with a accuracy rate of 70%, but only 63% in more complex searches, ranking third from the bottom. On average, each search returns about 7000 token; although the information is rich, this increases the model's processing costs. Completing one evaluation task costs an average of about $0.032, of which the search fee is $0.0085, with model invocation costs accounting for approximately 73% of the total cost.