# DeepSeek documents experimental vision model for its API

_Friday, August 21, 2026 at 6:33 AM EDT · AI, Products · Latest · Tier 2 — Notable_

![DeepSeek documents experimental vision model for its API — Primary](https://api-docs.deepseek.com/img/deepseek-social-card.jpeg)

DeepSeek's API documentation says its experimental deepseek-v4-flash-vision-exp model accepts image and text inputs through OpenAI-compatible Chat Completions and Responses APIs, as well as an Anthropic-compatible endpoint. Developers can provide images as base64 data, public URLs or uploaded files. The documentation sets a 48 MiB request-body limit for inline images, a 32 MiB limit for URL-downloaded images, and says images are resized before inference to a maximum of 384 image tokens per image.

## Sources

- [api-docs.deepseek.com](https://api-docs.deepseek.com/guides/vision/)

---
Canonical: https://techandbusiness.org/newswire/Z4emoJV0Q5o1oamn73lNSM
Retrieved: 2026-08-21T13:04:21.990Z
Publisher: Tech & Business (techandbusiness.org)