DeepSeek launches vision model deepseek-v4-flash-vision-exp
Original: DeepSeek-v4-flash-vision-exp
Why This Matters
DeepSeek expanding into multimodal APIs signals growing competition in the vision-language model space.
DeepSeek has released deepseek-v4-flash-vision-exp, a multimodal model accepting images alongside text via its API. It supports JPEG, PNG, GIF, and WebP formats and is compatible with OpenAI's Chat Completions API format.
DeepSeek has officially documented its new vision-capable model, deepseek-v4-flash-vision-exp, in its API documentation. The model accepts both image and text inputs, enabling use cases such as image description, screenshot text extraction, and chart analysis.
Developers can provide images through three methods: (1) Base64-encoded inline images embedded as data URLs, subject to a 48 MiB request body limit; (2) External HTTP/HTTPS URLs up to 8,192 characters, with image files up to 32 MiB and a 60-second download timeout; and (3) References to files pre-uploaded via the Files API, which supports files up to 64 MiB and is recommended for reuse across multiple requests.
The model uses the OpenAI-compatible Chat Completions API format at base URL https://api.deepseek.com, and image format is detected from actual file content rather than filename or declared MIME type. The same image input methods are also supported through DeepSeek's Responses API.