VK has started to implement a multimodal AI model in search

The VK social network has begun to introduce an artificial intelligence (AI) model into the search for its services, which is able to search for results simultaneously by text, image, sound and video sequence. This was reported on February 19 in the press service of the social network.

"VK has started to introduce visual language models (VLM), an artificial intelligence that simultaneously analyzes text, images, sound and video sequences, into the search for its products. The technology is already working in VK Video and will gradually appear in other services that have search engines," the company's website says.

It clarifies that when receiving a search query from a user, the model takes into account the name, description and content of the already uploaded content and gives a more accurate response. In addition, the new model automatically generates tables in which this data is stored.

The company noted that in the future, search engines will rely on the semantic meaning of the query. The introduction of such a model will accelerate the development of new search technologies fivefold and make the results more personalized.

"For example, the system will understand that the user often chooses videos with a certain style of editing and color correction. Or, more precisely, to recognize hybrid queries where text and visual characteristics are combined, for example, "a blog from Istanbul with views of the Bosphorus," the press service explained.

According to data from the analytical company Mediascope, the VK Video video service became the most popular video hosting service in terms of audience reach in Russia in January 2026. The total number of VK Video users per month reached 82.8 million people. According to analysts, 65.9 million people visited YouTube during the same period, and 47.7 million visited Rutube.

All important news is on the Izvestia channel in the MAX messenger.