August 22, yesterdayDeepSeek Online experimentalvisual understanding model DeepSeek-V4-Flash-Vision-Exp and open to API platform. Users can call through the deepseek-v4-flash-vision-exp model name; officially, its pure text, Agent, reasoning and knowledge skills are essentially the same as the V4-Flash official version, while adding image understanding capabilities。
The model is higher than Opus 4.8 in Argentinas' Last Exam and ZeroBench and slightly lower in ApexBench and Chartography in the official 4 multimodular Agent assessments. The above results are derived from the manufacturer ' s tests, and the Agent framework and mission settings used in the tests are given by the official authorities。
API supports the three call formats Chat Completions, Messages and Reponses, which allow for the entry of images in base64, external URLs and Files API. One picture is converted to a maximum of 384 tokens at the same price as V4-Flash. The simultaneous open Files API is free of storage charges and can be re-used via file\_id when the same picture is uploaded。