DeepSeek Launches Vision Model to Challenge Anthropic

Chinese artificial-intelligence startup DeepSeek is extending its challenge to Western model developers from text-based reasoning into visual understanding. Multimodal systems can interpret text alongside images and interface screenshots, a capability increasingly central to AI agents that navigate software, process documents and complete complex workflows. The move brings DeepSeek into more direct competition with premium offerings from Anthropic and other leading AI developers.

DeepSeek unveiled the experimental V4-Flash-Vision-Exp model with support for image and screenshot analysis. The company said performance improved sharply on multimodal agent benchmarks and approached Anthropic’s high-end Claude Opus 4.8, though it did not disclose scores or detailed comparisons. As of Aug. 24, 2026, the model was available to developers through DeepSeek’s API platform, allowing customers to begin testing its visual capabilities in applications.