DeepSeek has released DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal model that extends its text-only V4-Flash with image and screenshot understanding, and reports agent performance close to Anthropic’s Claude Opus-4.8 across several benchmarks. Benchmark tables show the model beating Opus-4.8 on three of eleven tests, while trailing on the rest, sometimes by wide margins. DeepSeek stresses that comparisons do not include Anthropic’s newer Claude Opus 5, whose performance against the model remains untested. Because DeepSeek’s systems are significantly cheaper to run than Anthropic’s, small score differences could still appeal to enterprise buyers, though a substantial deficit on repository-scale tasks may limit adoption until independent evaluations and Opus 5 comparisons arrive.
This update represents a notable development in the Ai sector. Organizations and founders tracking this space should evaluate potential strategic and technical implications on their operations.