DeepSeek says its AI agent app Harness hit 1.54 million daily users in under three months

Chinese AI company DeepSeek disclosed that its open-source agent application, Harness, reached 1.54 million daily active users on November 4, less than three months after launch. Bloomberg reported that those users consume about 92 trillion tokens a day between them, roughly 60 million per user per day, a sign that agents are being handed long, multi-step jobs rather than simple chat queries.

Users up more than fivefold in five weeks

Harness version 0.1 went live on August 13, with a desktop version following on September 29. According to figures disclosed by the DeepSeek team, daily active users stood at about 271,000 at the end of September and about 429,000 on October 7, before climbing to 1.54 million on November 4, more than fivefold growth in about five weeks. The team also said the ecosystem now has around 35,000 third-party plugins, and about 60 percent of active users have built or installed at least one.

Plugin marketplace planned; figures not independently verified

DeepSeek said it is preparing an official plugin marketplace and a support programme for plugin creators, to make it easier for developers to publish plugins and for users to find them, but gave no launch date. The figures come from DeepSeek's own disclosure and were not independently verified in Bloomberg's report; they also count only users connecting to DeepSeek's servers through its official application programming interface, not developers running the open-source code on their own infrastructure.

Context: from reasoning models to agent infrastructure

DeepSeek, headquartered in Hangzhou, shook Silicon Valley last year with its low-cost reasoning model R1. The company formed its agent Harness team in March 2026 and published a technical article on October 31, Super-agent Design Exploration, setting out its direction for models that carry out multi-step tasks on their own. Harness is that infrastructure: software that lets a model use tools, manage tasks and see work through, rather than simply answering questions. Bloomberg noted that DeepSeek is challenging Anthropic and OpenAI with open-source agent applications.