DeepSeek V4 Flash gains vision and agentic actions in new test model

DeepSeek has released DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal model that can interpret images and screenshots and act on what it sees. The company says it keeps the text performance of V4 Flash while approaching Anthropic’s Opus 4.8 on multimodal agent tests.
Source

👉@sysadminoff

https://4sysops.com/archives/deepseek-v4-flash-gains-vision-and-agentic-actions-in-new-test-model/