DeepSeek visual model multi-modular Agent approaches Opus 4.8

DeepSeek published the first multi-modular agent scores of v4-flash-vision-exp. In four multi-modit agnent assessments, it was two-to-two-by-two with Opus 4.8: 27.3 against 25.7, 35.0 against Zerobench, 36.5 against 39.4, and 63.3 against 65.0 against Opus 4.8. Compared to v4-flash-0731, the visual version rose from 26.2 to 36.5 in apexbench and from 25.2 to 27.3 in agents' last example. It needs to be noted that this set of results comes from the official DeepSeek self-assessment and is not a third-party stand-alone list。