On August 21, 2026, DeepSeek proudly announced the official release of its experimental multimodal visual understanding model, the DeepSeek-V4-Flash-Vision-Exp, on the DeepSeek API platform. To utilize this cutting-edge model, users simply need to configure their settings with model="deepseek-v4-flash-vision-exp". When it comes to text-based functionalities, including Agent operations, reasoning tasks, and leveraging world knowledge, this model performs on par with the standard DeepSeek-V4-Flash version. However, in scenarios demanding visual comprehension, as evaluated by the Agent Benchmark tests, it has demonstrated a marked enhancement over its predecessor, DeepSeek-V4-Flash. Its multimodal Agent capabilities are now nearing the sophisticated level achieved by Opus-4.8.