DeepSeek, which offers AI models through consumer chat products, developer APIs and downloadable software, has reportedly raised about $15 billion. Bloomberg reported the financing, citing people familiar with the matter; Reuters carried the report and said it couldn't independently verify it. Tencent and battery maker CATL were among the biggest investors, according to Bloomberg.
The report marks a substantive update to an unfinished round Captables covered earlier this month. On October 6, Reuters said DeepSeek expected to complete the fundraising later in October, while Bloomberg said the final amount could reach roughly the level now reported. The new account says the capital has been raised, not merely targeted.
The hosted products reach two main user groups. Individuals use DeepSeek through its website and app. Software teams send text—and, through the Flash model, images—to the hosted API, then receive generated responses inside their own applications. The service supports tool calls, structured output and thinking or non-thinking modes, letting developers add model capabilities without operating the underlying infrastructure.
DeepSeek also distributes model weights, giving teams a route to run the software on infrastructure they control. Its October 17 V4.1-Flash-1017 release was made available through the website, app and API, as well as for download through Hugging Face and ModelScope. DeepSeek describes the model as optimized for end-to-end software projects, with demonstrations including game creation and three-dimensional modeling.
The hosted service gives DeepSeek a direct revenue model. It bills developers according to the input and output tokens their applications consume, with separate per-request charges for search and webpage fetching. That puts a paid managed product alongside downloadable model weights: teams can either operate the model themselves or pay DeepSeek to handle inference through its API.