Chinese AI startup DeepSeek says its new multimodal model outperforms its prior version and rivals Anthropic’s Opus 4.8 in benchmarks.
DeepSeek announced the release of its experimental multimodal AI model, V4-Flash-Vision-Exp, which it claims surpasses its previous V4-Flash model on multimodal agent benchmarks. The company stated the new model performs close to Anthropic’s Opus 4.8, a leading AI model in the sector.
The startup did not provide specific performance metrics but indicated the model matches DeepSeek-V4-Flash on certain tasks. Prior versions of DeepSeek’s models have gained attention for their cost-efficiency and competitive performance against U.S. counterparts.
No immediate market reaction was reported following the announcement.