Less than three months after launch, DeepSeek’s open-source AI agent platform Harness has reportedly crossed 1.54 million daily active users while consuming 92 trillion tokens daily. That volume works out to roughly 60 million tokens per user on average and marks a sharp move away from simple chatbot exchanges toward full agentic workflows that run code, browse, and call tools on their own. The surge is already squeezing proprietary players like OpenAI and Anthropic and exposing fresh limits in global inference capacity.
Quick Take
- The milestone: DeepSeek Harness reached 1.54 million daily active users and 5.48 million monthly users by early November, a massive leap from just 271,000 daily users in late September.
- The token burn: Processing 92 trillion tokens daily mathematically exposes the high-compute reality of autonomous AI agents compared to standard prompt-and-response interfaces.
- The ecosystem play: With 35,000 user-built plugins already active and a marketplace looming, DeepSeek is aggressively decentralizing and open-sourcing AI agent development.
- The hidden caveat: These staggering figures only account for traffic routed through DeepSeek’s official API and servers, meaning the platform’s true global footprint—including self-hosted enterprise nodes—is likely much larger.
What happened
Launched on August 13, DeepSeek’s open-source agent application Harness has grown at an unusual pace. Recent reports show the platform hit 1.54 million daily active users by November 4, up sharply from about 271,000 DAUs at the end of September, a jump that lined up with the release of its desktop client.
Engagement runs deeper than user counts alone. Harness now handles around 92 trillion tokens each day, or roughly 60 million tokens per daily user. That scale comes from agents that write and execute code, scan files, run ongoing web research, and trigger external tools without constant human input.
The platform is built to be extended. It already has about 35,000 user-created plugins, and developers can replace even core features with their own code. DeepSeek has said it will open a plugin marketplace to support and monetize this growing ecosystem.
Why it matters now
The Harness numbers give the clearest view so far of the compute costs that will come with widespread agent use. Sixty million tokens per user per day is a step change from ordinary chat. Agents that loop through code, web calls, and tool actions drive token use far higher than before. If this pattern spreads, inference demand will test data-center capacity, power grids, and GPU supply, pushing the industry toward cheaper inference designs to stay viable.
On the competitive side, the results land squarely on OpenAI and Anthropic. DeepSeek pairs low-cost models with an open plugin system that lets developers customize deeply, showing that many teams prefer transparent pricing and open control over closed platforms.
The 92 trillion daily tokens figure covers only traffic on DeepSeek’s own servers and APIs. Because the models are open-source, the real footprint includes self-hosted enterprise setups and regional instances that never reach the public numbers. That gap means the platform’s overall reach and infrastructure load are likely understated.
Who is most affected
The quick rise of Harness changes the picture for developers, model builders, and infrastructure providers alike. Developers now have a flexible, low-cost place to build agents, while established labs face pressure to defend higher prices against a credible open-source option. At the same time, chip makers and data-center operators must plan for heavier, always-on inference loads as these systems move into regular enterprise use.
| Stakeholder | Impact | Why |
|---|---|---|
| Proprietary AI Labs (OpenAI, Anthropic) | High | Must defend enterprise pricing and walled-garden agent tools against a hyper-growing, open-source alternative with comparable traction. |
| AI Developers & Founders | Positive | Access to a highly extensible platform with 35K+ plugins and low-cost inference drastically lowers the barrier to building autonomous agents. |
| AI Infrastructure Providers | High | 92 trillion daily tokens from just 1.54M users signals an incoming tidal wave of continuous inference compute and network demand. |
About the analysis
Researched and written with AI assistance under the editorial responsibility of the i10X News team. The title image is an AI-generated illustration. Read our AI Use Policy. Spotted an error? Request a correction.
This is an independent, research-based i10x analysis designed for AI founders, CTOs, and infrastructure strategists, synthesizing global technology reporting and competitive market data to track the evolution of agentic platforms.
i10x Perspective
The real story is not another app crossing a million users but a direct test of the compute load that agentic AI will place on the system. When daily use settles at 60 million tokens per person, older pricing models and capacity forecasts stop working. DeepSeek shows that low-cost inference plus open extensibility can match the resources of big Silicon Valley labs, yet the surge in demand that follows may become the tighter constraint in the race ahead.
