DeepSeek launches vision model rivaling Anthropic’s Opus 4.8

DeepSeek released an experimental multimodal artificial intelligence model on Friday, August 21, 2026, that adds image understanding to its existing text-focused V4 Flash model. The Chinese AI startup stated the new model, named DeepSeek V4 Flash Vision Exp, approaches the performance of Anthropic's Opus 4.8 on specific multimodal agent tasks, marking a significant step in its efforts to compete with leading U.S. AI systems.

The new model preserves the text-based reasoning and world knowledge capabilities of the V4 Flash foundation model but can now interpret visual inputs such as images and screenshots. This allows it to perform tasks like describing visual content, extracting text from images, and analyzing diagrams, according to the company. DeepSeek is positioning the model for "agent-based applications," designed to work with various agent frameworks to combine visual understanding with tool use.

In benchmarks published by DeepSeek, V4 Flash Vision Exp performed close to Anthropic's Opus 4.8 on several tests. The model outperformed Opus 4.8 on three of the eleven reported benchmarks: Agents Last Exam, ZeroBench, and DeepSWE. However, it trailed on the majority, with the...