DeepSeek says its experimental multimodal model V4-Flash-Vision-Exp is now live on the DeepSeek API Platform

DeepSeek product releases have become a recurring market event rather than routine corporate news: past model launches from the lab have repriced assumptions about the cost curve of frontier AI capability, and on previous occasions that repricing transmitted well beyond China, hitting the global capex and compute narrative and the peer set of AI-linked semiconductor and infrastructure names. The immediate question is capability and cost relative to incumbent frontier models, since that is the axis along which prior DeepSeek announcements moved markets, rather than the API availability itself. A multimodal release widens the comparison set from text models to vision-capable systems, which is where the larger labs have concentrated recent effort. The tell in past episodes has been independent benchmark verification and developer adoption over the following days, not the launch copy. Absent verified performance and pricing detail, the headline alone carries limited signal; the follow-on coverage and any commentary from competing labs are what have historically determined whether a release fades or drives a sustained repricing.