DeepSeek

DeepSeek is an AI company founded in 2023 in Hangzhou, China, by Liang Wenfeng. What's unusual is that its parent company is a hedge fund — the fund provides the money, and a small research team builds the models.

The world took notice because of cost. DeepSeek released models that came close to the performance of top US models at a fraction — literally a few percent — of the price American companies spend. On top of that, it released the full internals of its models openly, meaning anyone can download them and run them on their own servers.

There are two common misconceptions. One is that DeepSeek is a Chinese government agency or a subsidiary of a big conglomerate — in reality, it's an independent research team funded by a hedge fund. The other is the assumption that "cheap means worse performance." The company itself acknowledges it still trails the very top US models slightly, but that gap has narrowed to a matter of months.

As of August 2026, its flagship is the V4 generation. The officially released V4 Pro lets users adjust reasoning depth across three levels, while the lightweight V4 Flash is small enough that some people run it on personal PCs. DeepSeek is the source of the "price war" that pushed OpenAI and Anthropic to cut their own prices in response, and it has also formed a dedicated team (the harness team) building coding tools aimed at competing with Claude Code.

"The DeepSeek shock" — this phrase is usually about cost rather than raw performance. When a model can do the same job for much less money, it puts pressure on the price of more expensive models.