models
Artem Vysotsky
DeepSeek V4 Flash Vision Exp Is on Writingmate: Testing Image Understanding on a 1M-Token Model
DeepSeek shipped its first vision-capable Flash model on August 21, 2026. I ran it through the same broken-function, agentic, and long-context tests I use on every release, then added a round of screenshots, charts, and scanned documents to see where "experimental" actually shows up.
Read Article


