DeepSeek V4 Flash Vision Experimental is now live on Vercel AI Gateway under the model ID deepseek/deepseek-v4-flash-vision-exp. It accepts images alongside text in a single request, supporting JPEG, PNG, GIF, and WebP. File type is detected from the file's own bytes, not the declared mediaType, so mislabeled files still process correctly.
Tool use, reasoning, and caching carry over from the non-vision version unchanged. For coding agent workflows, running vercel ai-gateway coding-agents setup connects it to Claude Code, Codex, OpenCode, Cursor, and Pi. The -exp suffix in the model ID signals an experimental release: behavior can change, and Vercel explicitly recommends keeping a fallback model on any production path.
Pricing mirrors provider rates with no markup and no platform fee, including BYOK requests. The original changelog is worth reading for the exact image input format spec and the model playground link, especially if you are evaluating this against other vision models already on the gateway.
[READ ORIGINAL →]