DeepSeek launches an experimental multimodal model to rival Anthropic

DeepSeek unveiled an experimental multimodal model, DeepSeek-V4-Flash-Vision-Exp, which extends its text-only V4-Flash system with image and screenshot reading capabilities and is now accessible via the company’s API platform. The model’s agent performance is reported to be close to Anthropic’s Opus-4.8, and according to DeepSeek’s own benchmark table it surpasses Opus-4.8 on three of eleven evaluated tasks.