TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
7x24hNews

DeepSeek launches multimodal vision understanding model V4-Flash-Vision-Exp, enabling multimodal API services.

2026.08.21 - 09:30
Share

TechFlow reports that on August 21, DeepSeek announced today that its brand-new multimodal visual understanding model, DeepSeek-V4-Flash-Vision-Exp, is now live on the DeepSeek API platform. The model is experimental, and users can invoke it by setting model="deepseek-v4-flash-vision-exp". Its pure text capabilities are roughly on par with the official version of DeepSeek-V4-Flash, showing significant improvements in Agent benchmark tests requiring visual understanding, with its multimodal Agent performance approaching that of Opus-4.8.

The model supports three call formats: Chat Completions, Messages, and Responses, enabling integration with various Agent tools. It supports mixed text-and-image input, as well as three methods for submitting images: inline base64, external URLs, and the Files API. DeepSeek has also opened up the Files API, allowing users to upload images and reference them via file_id to reduce repeated uploads and minimize request bandwidth usage.

Add to Favorites
Share to Social Media