Processing 92 trillion tokens every day, DeepSeek's open-source artificial intelligence agent platform, Harness, has reached 1.54 million daily active users in under three months.
DeepSeek reported the user milestone as of November 4, 2026, less than three months after launching the platform in August. As Tian Feng, president of the Fast Think Institute, noted, China's AI ecosystem shows strong usage patterns and climbing user engagement, explaining that genuine competitiveness depends entirely on whether AI models integrate into daily workflows and generate real utility. Tian described these agent tools as acting like a "digital employee," yet he cautioned that higher usage alone does not guarantee business success or profit.
However, massive technical engagement and high generation volumes do not directly resolve the challenge of infrastructure costs, leaving the commercial durability of the platform open. The fundamental question of whether DeepSeek's infrastructure can sustain the costs of maintaining 92 trillion daily tokens remains unresolved by the current metrics.
The Architecture of Open Extensions
The DeepSeek Harness platform launched in August 2026 as an open-source repository. Unlike standard conversational models designed for simple question-and-answer interactions, Harness operates as an environment where models connect applications and carry out multi-step tasks through modular extensions.
That architectural flexibility allows engineers to add or swap functionality without altering the underlying core codebase. The network now hosts nearly 35,000 third-party plugins, indicating that external contributors are actively building out capabilities beyond DeepSeek's baseline release.
"These tools now handle longer tasks and act like a 'digital employee.'"
DeepSeek stated that approximately 60% of active users have either built or installed at least one plugin. This behavioral pattern separates the runtime from standard chat interfaces, demonstrating that users are relying on the framework to execute automated operational functions rather than passive query processing.
Monetization Pressures Behind Surging Usage
To formalize this developer activity, DeepSeek plans to introduce an official plugin marketplace where third-party contributors can publish software tools while end users discover applications. The company also intends to support plugin builders by offering token rewards, technical discussions, and promotional visibility.
Industry observers consider this approach an effort that could bolster DeepSeek's competitive position against proprietary developers such as OpenAI and Anthropic. Yet, as Tian Feng's assessment makes clear, genuine competitiveness depends entirely on integration into daily workflows, and higher usage alone does not guarantee business profit.
DeepSeek faces the operational challenge of sustaining developer interest, keeping users active, and delivering dependable applications. Meanwhile, underlying model consumption across the ecosystem continues to climb: DeepSeek's V4.1-Flash-1017 model logged 56 trillion tokens in weekly usage between November 2 and November 8, 2026, marking a 17% jump from the previous week.
Against the field
Scaling an agentic runtime to over a million active users threatens to trigger an unsustainable infrastructure bill as token consumption climbs. In mid-August the project gained over 140,000 GitHub stars right away, but developer excitement has now collided with brute-force computing costs. With almost 35,000 community plugins triggering complex multi-step pipelines, an open ecosystem risks turning into an expensive compute sink rather than a profitable product. Whether the planned plugin marketplace can monetize this traffic before compute overhead exhausts company resources remains the key operational test.
Our stories: DeepSeek Harness: Open-Source Runtime for AI Agents (August 17).