DeepSeek Ships Its First Vision Model: V4-Flash-Vision-Exp

DeepSeek added an experimental multimodal variant of V4-Flash to its API, matching the text model's agent and reasoning scores while adding a major leap in visual understanding, closing in on Opus-4.8 on multimodal agent tasks.