88Z AI, 1 September — The first multimodal model in the DeepSeek V4 series, DeepSeek-V4-Flash-Vision-Exp, has been open-sourced under the MIT license. It natively supports image inputs in JPEG, PNG, GIF, and WebP formats, offering capabilities such as image description, text recognition, and chart analysis. The model shows significant improvement in visual Agent benchmarks, with multimodal Agent capabilities approaching Opus-4.8, while pure text task performance matches the official V4-Flash release.
At a glance
- DeepSeek-V4-Flash-Vision-Exp is open-sourced under the MIT license, natively supporting multiple image formats and visual analysis features.
- The model's multimodal Agent capabilities approach Opus-4.8, while pure text performance matches the official V4-Flash release.
The update
DeepSeek has officially launched DeepSeek-V4-Flash-Vision-Exp, the first multimodal model in its V4 series, releasing it under the MIT license. The model is designed to natively support various common image formats, including JPEG, PNG, GIF, and WebP, allowing for direct processing of image inputs. Its core features include detailed descriptions of image content, text recognition within scenes, and in-depth analysis of various chart data. In terms of performance, the model has achieved significant improvements in visual Agent benchmarks. Its comprehensive multimodal Agent capabilities are now very close to the level of Opus-4.8. Meanwhile, when handling pure text tasks, the performance of this experimental model remains consistent with the official V4-Flash release, ensuring the stability of basic language capabilities.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- IT之家Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.