DeepSeek has disclosed usage figures for Harness, its AI agent platform, for the first time — and the numbers show one of the fastest adoption curves in the current agent-coding boom. The platform counted 1.54 million daily active users as of November 4, consuming about 92 trillion tokens a day, DeepSeek said on November 5.

The disclosure, reported by Chinese financial outlet Yicai and picked up by TechInAsia, marks the first time DeepSeek has put public numbers behind Harness, the desktop agent workspace where AI agents write code, run shell commands and produce documents on a user's own machine. The figures count users of the platform's official application programming interface.

A fivefold jump in five weeks

Harness has grown at a striking pace since its desktop launch. Daily active users, measured as a seven-day average, stood at just 271,000 on September 30, climbed to 429,000 by October 7, and reached 1.54 million by November 4 — adding more than a million users in a single month, and more than five times the late-September level.

The token figures are equally large. At roughly 92 trillion tokens a day, Harness is now processing a volume of AI work that puts it among the most heavily used agent platforms anywhere — a scale that also helps explain why DeepSeek has been racing to expand its model lineup, including the DeepSeek V4 Pro and Flash models that power the platform.

DeepSeek has not said how many of those users are paying customers, and the figures are company-reported rather than independently audited. But the direction is unambiguous: the agent platform has gone from a niche developer tool to a mass-market product in about five weeks.

Plugins are doing the heavy lifting

Much of that growth appears to be driven by Harness's plugin ecosystem. The number of third-party plugins has risen to about 35,000 since launch, DeepSeek said, and roughly 60 percent of active users have created or installed at least one plugin.

Plugins extend Harness beyond its built-in abilities, adding integrations, workflows and tools that users build for themselves and share with others. DeepSeek said it is preparing a support programme for plugin authors, built around token incentives, technical exchange and promotion, alongside an official plugin marketplace.

The approach mirrors the strategy that has made other agent platforms sticky: once users have invested in custom tools and workflows, they are far less likely to leave. For DeepSeek, whose models are also available through rival platforms, owning the workspace where the work happens matters as much as the model itself.

The agent race keeps accelerating

Harness's surge lands in the middle of a crowded fight over the agent-coding workspace. Anthropic's Claude Code, OpenAI's Codex and a wave of startups are all competing to become the place where developers hand real work to AI agents — and usage, not benchmarks, is becoming the scoreboard that matters.

DeepSeek's disclosure also underlines how quickly the centre of gravity is shifting. A Chinese lab best known for cheap, open models is now reporting usage numbers that rival the biggest Western agent platforms, barely a month after Harness's desktop launch.

Whether that pace holds is an open question. Growth driven by a launch surge and a plugin gold rush can cool quickly, and DeepSeek will need to convert free usage into durable revenue. For now, though, Harness has given DeepSeek something it did not have a month ago: a fast-growing platform of its own, with more than a million daily users inside it.