DeepSeek Harness reaches 1.54 million daily active users

By Tom Whitfield

How fast has DeepSeek Harness grown?

DeepSeek Harness reached 1.54 million daily active users on November 4, according to a post on X from the Hangzhou-based Chinese AI startup.

The open-source agent app launched on August 13. DeepSeek said its audience had grown nearly six times in a little more than a month. At the end of September, shortly after a desktop release, Harness had about 271,000 daily users. DeepSeek derived that estimate from the number of people connecting to its own servers.

Harness consumes approximately 92 trillion tokens per day. DeepSeek said that equated to roughly 60 million tokens for each active user every day.

The startup is betting on infrastructure for AI agents, software that lets artificial intelligence act autonomously on a user's behalf.

DeepSeek's growth curve has moved from hundreds of thousands to more than 1.5 million daily users.

What does DeepSeek plan for plugins?

DeepSeek plans to launch an official marketplace where developers can publish plugins and users can find them. The company also plans a support program for plugin developers. It will include token grants, technical assistance and opportunities for promotion, and DeepSeek disclosed no further details.

The initiatives reflect a broader shift in artificial intelligence, with competition for users increasingly centered on ecosystems and integrations rather than model performance alone, DeepSeek said.

In DeepSeek's plan, the marketplace extends the company beyond the model and into the developer tools around it.

How does the open-agent architecture work?

DeepSeek is developing Harness on the premise that “tudo é um plugin,” meaning everything is a plugin. A plugin adds one specific capability to a program without modifying the program itself, expanding what the software can do.

The startup counts about 35,000 plugins. Sixty percent of users have created at least one, although DeepSeek said the sophistication of those tools varies.

DeepSeek positions Harness as an open, developer-led ecosystem for agents. DeepSeek aims to challenge OpenAI and Anthropic on model performance and cost, as well as on the variety of applications built around its AI technology.

DeepSeek believes third-party applications and specialist tools will drive adoption of its open agent infrastructure. DeepSeek expects those tools to become a decisive advantage over proprietary Silicon Valley technologies as AI moves beyond chatbots.

The 35,000 existing plugins give DeepSeek a base of extensions before the marketplace opens.

How much computing capacity do agents require?

DeepSeek's reported 92 trillion daily tokens provide one of the first available indications of how agents sharply increase computing demand. DeepSeek said agents require much more processing power than chatbots because they can operate for hours while reading or writing code, using other tools and browsing the internet with little human intervention.

Agents execute multistep tasks on users' behalf, including writing and modifying code, analyzing files, conducting research and using other tools, with minimal human supervision, DeepSeek said.

The startup surprised Silicon Valley last year with R1, a low-cost reasoning model. Harness carries DeepSeek's open-source positioning into its chosen next frontier: infrastructure for autonomous agents.

The next test for DeepSeek is whether plugins and developer support turn heavy agent use into a working ecosystem.