DeepSeek Launches Experimental Multimodal API Model V4-Flash-Vision-Exp
DeepSeek has launched DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal API model that supports image inputs alongside text. The model allows users to perform tasks like image description, text recognition (OCR), and chart analysis using formats like JPEG, PNG, GIF, and WebP. This release significantly enhances DeepSeek's API capabilities by adding vision support, enabling developers to build more versatile multimodal AI agents. Its vision-based agent performance reportedly approaches high-end models like Opus-4.8, offering a cost-effective alternative for complex visual tasks. The model is billed at the same rate as V4-Flash, with each image consuming a maximum of 384 tokens. Additionally, DeepSeek introduced a free Files API that allows users to upload images once and reference them via file_id to save request bandwidth.
## BACKGROUND
Multimodal AI models can process multiple types of data inputs, such as text and images, simultaneously. AI agents leverage these models to perform autonomous tasks, and vision-based agent benchmarks (like VisualAgentBench) evaluate their ability to navigate graphical user interfaces (GUIs) or solve visual reasoning problems.