The DeepSeek-V4-Flash-Vision-Exp model has been officially open-sourced.

DeepSeek-V4-Flash-Vision-Exp, the first multimodal model in the DeepSeek V4 series, has been open-sourced and released under the MIT license. The model natively supports image input, enabling image description, text recognition, and chart analysis, and supports JPEG/PNG/GIF/WebP formats. It demonstrates significant performance improvements in visual agent benchmarks, with multimodal agent capabilities approaching those of Opus-4.8, and its performance on plain text tasks matching that of the official V4-Flash release.