DeepSeek, a Hangzhou-based AI developer, said on Aug. 21 it had released an experimental model called DeepSeek-V4-Flash-Vision-Exp through its API, adding image and screenshot analysis to its V4 Flash text model as Chinese companies expand multimodal AI offerings.

DeepSeek said the model retains V4 Flash’s text capabilities, including reasoning, agents, and world knowledge.

It also said the model’s multimodal agentic performance, which means carrying out tasks with limited oversight, is “close to” Anthropic PBC’s Opus 4.8.

The company previously released multimodal and visual models under its DeepSeek-VL family, but said the new release brings image interpretation and task execution to its latest model series.

The release did not show specific pricing or input limits for the model, and did not specify separate API retention or image-handling terms for screenshot uploads.

🔗 Source: Bloomberg