China’s DeepSeek has introduced an experimental model equipped with visual comprehension, claiming its new system closely rivals the capabilities of an advanced model developed by Claude-maker Anthropic. The new release represents an upgraded multimodal build of the company’s flagship text-based V4 Flash architecture. It enables the system to process, evaluate and execute tasks based on visual inputs such as still photographs and screen captures.
Maintaining core strengths while adding vision
DeepSeek confirmed in a post on X that this experimental multimodal release preserves the foundational text benchmarks of DeepSeek-V4-Flash, including its reasoning capacity, broad factual knowledge, and autonomous agent features.DeepSeek stated that the model’s performance comes close to Anthropic PBC’s Opus 4.8 model across multimodal agentic evaluations which are tests designed to measure an AI system’s ability to complete multi-step tasks independently without continuous human intervention or prompting.
What DeepSeek claims
DeepSeek-V4-Flash-Vision-Exp is now live on the DeepSeek API Platform! 🚀🔹 This experimental multimodal model matches DeepSeek-V4-Flash on text capabilities—including agents, reasoning, and world knowledge.🔹 On multimodal agent benchmarks, V4-Flash-Vision-Exp makes a major leap over V4-Flash, bringing multimodal agent performance close to Opus-4.8.Try it with model=’deepseek-v4-flash-vision-exp’. DeepSeek Harness 0.1.1 was released today with out-of-the-box support for the new model.Multimodality unlocks more agent use cases. 👀V4-Flash-Vision-Exp works smoothly across agent frameworks, combining visual understanding with a wide range of tools to unlock more practical workflows.The tool, called DeepSeek-V4-Flash-Vision-Exp, provides visual processing and execution capabilities to DeepSeek’s frontline model family, and is available to developers through the company’s API platform.