skip to content
The Weighted Average

Wire

DeepSeek opens V4 Flash Vision API

DeepSeek launched the experimental V4-Flash-Vision-Exp API, adding image input to V4 Flash with a maximum of 384 billed tokens per image and a free Files API for reusing uploads, according to DeepSeek’s release note. The model supports Chat Completions, Messages, and Responses, and DeepSeek says its multimodal agent scores approach Opus-4.8. Builders can now test screenshot, chart, and visual-tool workflows without paying the bandwidth cost of repeated uploads, while the earlier V4 performance work already showed how much local serving changes the economics.