Chinese artificial intelligence startup DeepSeek has introduced multimodal capabilities to its flagship chatbot, allowing the system to process images and video in addition to text for the first time, according to the South China Morning Post.
The update brings DeepSeek in line with competing AI platforms that have already offered image and video understanding for some time. The feature is currently being rolled out to a limited group of users, the report said.
The move comes just days after the Hangzhou-based company released its latest flagship model, V4, which was accompanied by significant price reductions on its services.
A milestone for the company
DeepSeek's multimodal team leader Chen Xiaokang is cited by the South China Morning Post in connection with the announcement, which has been described informally within the company as the moment "the whale can now see" - a reference to the whale imagery associated with the DeepSeek brand.
Multimodal AI systems are considered a key benchmark in the industry, enabling applications that go beyond text-based queries to include tasks such as analyzing photographs, interpreting charts, and processing video content. Major competitors including OpenAI's ChatGPT and Google's Gemini have offered comparable features for an extended period.
Rapid expansion from a fast-moving competitor
DeepSeek drew widespread international attention earlier this year after releasing models that demonstrated performance competitive with leading Western AI systems, while reportedly being developed at a fraction of the cost. That announcement prompted significant discussion across the technology industry about the assumptions underpinning AI development expenditures.
The addition of multimodal support, combined with the recent release of V4 and accompanying price cuts, signals continued aggressive expansion from the startup as it works to close feature gaps with established players in the global AI market.
No specific timeline for a broader public rollout of the vision capabilities has been announced, according to available reporting.
