DeepSeek V4 Flash Vision Experimental now available on AI Gateway
DeepSeek V4 Flash Vision adds screenshot, image, and chart input to Vercel AI Gateway while retaining tool use, reasoning, and caching, but its experimental ID signals production risk.
The experimental **deepseek/deepseek-v4-flash-vision-exp** model now accepts text with images through AI Gateway. It supports **JPEG, PNG, GIF, and WebP**, while tool use, reasoning, and caching remain available.
Builders can route screenshots or charts into coding-agent tasks without changing gateways. File type is detected from the image bytes, so agents do not depend on a correct filename or declared media type.
The experimental **deepseek/deepseek-v4-flash-vision-exp** model now accepts text with images through AI Gateway. It supports **JPEG, PNG, GIF, and WebP**, while tool use, reasoning, and caching remain available. Builders can route screenshots or charts into coding-agent tasks without changing gateways. File type is detected from the image bytes, so agents do not depend on a correct filename or declared media type. The **-exp** suffix means behavior may change. Keep a fallback model on production paths and validate vision-dependent workflows rather than assuming stable outputs.
This adds an experimental vision route for coding agents already using AI Gateway, making screenshot and chart input available without a gateway change. It broadens model choice but does not settle selection: production use still requires workload validation and fallback routing because the model is explicitly unstable.