DeepSeek Harness reports 1.54 million daily users

DeepSeek Harness said it had 1.54 million daily active users as of November 4, counting users of its official API.[1] Bloomberg also reported that it reached this level less than three months after its August 13 debut.[2][4] For comparison, Bloomberg reported about 271,000 daily users at the end of September, after the desktop version was released, based on users connecting to the service’s servers.[2][4] The team also reported daily token consumption of about 92 trillion.[1] These figures describe usage; they do not establish how users apply the product at work.

The platform had about 35,000 third-party plugins, and about 60% of active users had created or installed at least one.[1] DeepSeek Harness describes its goal as becoming an operating system for the Agent era, with core capabilities and plugins for specialized needs.[1] Its “Everything is a Plugin” approach allows third-party plugins to replace built-in functions.[1] The team said it would improve plugin API stability and compatibility, strengthen security, and build an official marketplace to help developers upload plugins and users find them.[4] Developers have created plugins in areas including content creation, design, context management, and scientific research.[1]

DeepSeek Harness version 0.1 launched on August 13, and DeepSeek released its desktop version on September 29.[4] DeepSeek says the desktop product can work in the background, edit local files, and handle long, complex tasks.[5] It is available for Apple-silicon Macs running macOS 13 or later.[5] The team says it will support plugin authors through token incentives, technical exchanges, and recommendations, with program details to be refined in response to community needs and feedback.[4] It has also invited startups and enterprises to connect products, services, and specialist capabilities through plugins and explore partnerships.[4]

A report comparing DeepSeek Harness with OpenCode described token consumption as more than 20 times higher and a task taking 30 to 40 minutes, versus about 30 seconds on OpenCode.[1] The report explicitly said this reflected only individual users’ experiences.[1] That caveat matters when evaluating the figures: the evidence does not establish a standardized benchmark or show that the same results apply across tasks and users.[1] Teams considering the product can treat the comparison as a prompt to test their own workflows, rather than as a general performance guarantee.