What this is
DeepSeek launched the vision version deepseek-v4-flash-vision-exp on August 21, priced identically to the text-only Flash tier. For the first time, "seeing images" achieves the same cost as "reading text." Developers simply swap the model to the vision version in API calls (i.e., visual recognition, letting AI understand images), then upload images for content identification and Q&A. Official Chat, DeepSeek Harness, and cc-switch (a unified model configuration tool that lets Claude Code call different models) all support it, with measured response times around 4–13 seconds.
Industry view
The open-source camp has long had a weak spot in visual understanding—Claude Code still officially doesn't support image recognition, GPT-4o and Doubao carry closed-source multimodal pricing thresholds that aren't low, and an open-source vision solution that actually runs has barely existed. What DeepSeek delivers: a Flash-priced, seconds-fast, drop-in open-source vision model that hooks directly into the Claude Code CLI (command-line interface). "Affordable" and "usable" are true at the same time for the first time.
But we see points worth watching. First, this is an experimental model (the "exp" in the name stands for experimental)—stability and long-term availability need verification. Second, the vision capability is packed into a mid-tier slot like Flash, which can attract developer traffic, but whether it holds up under enterprise-grade high concurrency (many users calling simultaneously) is an open question. Third, for enterprises, the real decision factors for switching remain engineering metrics like latency and data compliance—a single tutorial won't solve that.
Impact on regular people
For enterprise IT: if your team already runs Claude Code or similar tools, you can try shifting image recognition tasks from GPT-4o to DeepSeek next month. At current pricing, recognition cost per million tokens may drop from several dollars to a few RMB cents.
For individual professionals: you can hook your own machine's Claude Code to DeepSeek's vision version to handle invoices, screenshots, and spreadsheets without opening a second tool. The prerequisite is willingness to tinker with command-line configuration, which is a non-trivial barrier for non-technical users.
For consumer markets: image recognition has long been standard in free consumer-facing (C-side) Chat apps. DeepSeek's move benefits the B-side (enterprise) and developer users more; ordinary consumers' daily usage will see limited impact.