The artificial intelligence agent platform DeepSeek Harness reached 1.54 million daily active users less than three months after its launch. The growth came accompanied by a consumption of approximately 92 trillion tokens per day, according to figures released by the Chinese company and reported by Bloomberg on Sunday (8).
The data, referring to November 4, consider users who access the service through DeepSeek's official API. The processing volume is equivalent to an average of approximately 60 million tokens per active user daily, although the company did not detail how this consumption is distributed among different activities.
Harness was launched on August 13 and received a version for computers on September 29. At the time, the average daily active users, calculated over seven days, was around 271,000. In just over a month, the registered audience became almost six times larger.
The platform allows executing tasks in multiple steps with AI agents, including programming, document analysis, research and use of external tools. Unlike a conventional conversation with a chatbot, these operations can involve prolonged sessions, with successive calls to the models to interpret information and execute actions.
Ecosystem already brings together 35,000 plugins
In addition to audience growth, DeepSeek reported that Harness already has approximately 35,000 plugins developed by third parties. About 60% of active users have created or installed at least one of these extensions, although the company acknowledges significant differences in the maturity of the available tools.
The system uses an architecture in which practically all functionalities are implemented by plugins. This allows developers to replace even features originally integrated into the platform with their own solutions, adapted to different activities.
To expand this ecosystem, DeepSeek announced plans to create an official plugin marketplace and a developer incentive program, with the granting of token credits, technical assistance and promotional opportunities. The company also intends to improve the stability, compatibility and security of extensions, but has not yet set a date for the launch of these initiatives.
The released numbers offer a sense of the processing involved in operating autonomous agents. However, since the metrics were provided by DeepSeek itself, without a disclosed independent audit, they do not allow determining in isolation the computational cost of each task nor establishing a direct comparison with the consumption of traditional chatbots.
