DeepSeek says its open-source AI agent application, Harness, reached 1.54 million daily active users by 4 November, less than three months after it launched, as the Chinese startup steps up its challenge to OpenAI and Anthropic in the market for autonomous AI assistants.

The figures were disclosed by the company on Sunday, 8 November, and reported by Bloomberg. At the end of September, Harness had roughly 271,000 daily users, so the count rose by more than a million in a little over a month. The tally counts only people connecting to DeepSeek's own servers, and the figures are the company's own disclosures rather than independently verified adoption or usage.

Harness launched on 13 August and a desktop version followed in late September. Unlike a chatbot that answers one question at a time, the app is designed to carry out multi-step tasks with limited human supervision, such as writing and modifying code, analysing files and conducting research, using specialised tools on the user's behalf.

Heavy token use points to agent workloads

DeepSeek also reported that Harness processes about 92 trillion tokens a day, which works out at roughly 60 million tokens per active user each day. Tokens are the units of information processed by AI models, and agent workflows can require repeated model calls as they read and write code, call tools and browse the web, so the figure underlines how much more computing power agents consume than short chatbot exchanges.

That appetite raises questions about operating costs as well as demand for processing capacity. DeepSeek has built its reputation on low-cost models, but running agents for hours at a time is a far heavier workload than answering single prompts, and the company has not disclosed what Harness costs it to operate.

Plugin marketplace planned

DeepSeek said Harness now has about 35,000 plugins, and that 60 percent of users have created at least one, although their complexity varies. Plugins add specialised functions to the application without changing its core software, and DeepSeek is positioning the ecosystem as the way an open-source platform can compete with proprietary offerings.

The startup plans an official marketplace where developers can distribute plugins, along with a developer support programme that would include AI usage credits, technical help and promotion. No launch date or further details have been disclosed for either initiative.

The growth highlights how competition in AI is shifting from model capabilities alone towards applications and developer ecosystems. Whether Harness can sustain its pace will depend on whether developers keep building for it, and whether its open approach can match the reach of the established platforms it is trying to challenge.