DeepSeek’s multimodal AI has entered testing

确认！DeepSeek多模态AI已经开测

DeepSeek researchers confirmed V4 vision mode is in gray testing, with an image-recognition mode on the homepage. A screenshot shows it identified drinks and cup types in a non-text-heavy image after 4 seconds. The post does not disclose rollout scope, API access, or pricing.

Why it matters: HKR-H/K/R all pass: DeepSeek’s V4 vision gray test is a real domestic flagship update with a concrete 4s sample. Score stays at 80 because access scope, API form, pricing, and benchmarks are not disclosed.