DeepSeek V4 Flash Vision Experimental now available on AI Gateway
Vercel’s AI Gateway now offers DeepSeek V4 Flash Vision Experimental, enabling multimodal inputs with images and text for tasks like image description, text extraction, and chart analysis.
Vercel has integrated DeepSeek V4 Flash Vision Experimental into its AI Gateway, allowing users to submit both images and text in a single prompt. The model supports JPEG, PNG, GIF, and WebP formats, processing images based on their byte content rather than file names. This experimental release, indicated by the -exp suffix in the model ID, is designed for testing and may undergo changes in behavior.
The model maintains compatibility with existing features such as tool use, reasoning, and caching. Vercel advises users to configure a fallback model for production environments due to the experimental nature of the release. The model ID for integration is deepseek/deepseek-v4-flash-vision-exp.
Developers can deploy the model in coding agents like Claude Code, Codex, OpenCode, Cursor, and Pi by running vercel ai-gateway coding-agents setup. Vercel’s AI Gateway connects these agents to the model without additional platform fees on inference, including Bring Your Own Key (BYOK) requests.
AI Gateway mirrors provider pricing with no markup, ensuring transparent costs for users. The platform provides access to a range of language models, with DeepSeek V4 Flash Vision Experimental available for testing in the model playground.