OpenRouter reports 7.3T agentic token usage by Aug. 10, up 14x since February

The machines are talking to themselves, and they’re getting increasingly chatty. OpenRouter, the AI model aggregation and routing platform, recorded 7.3 trillion tokens of agentic usage on a 7-day average as of August 10, a 14-fold increase from roughly 500 billion tokens on February 6.

Human token usage, by contrast, grew to 1.4 trillion tokens over the same period, a more pedestrian 2.8x increase.

Agents are eating the inference budget

On February 6, agentic usage first crossed above human levels on the platform. Six months later, agents are consuming more than five times what humans are.

Agentic workloads, where AI models execute complex multi-step tasks like coding, debugging, or automated research, consume roughly 13 to 15 times more tokens per individual request than a typical human interaction.

OpenRouter classifies its traffic using a weighted composite score that factors in tool calls, conversation turns, and timing patterns. This segmentation approach lets the platform distinguish between a human chatting with Claude and an automated pipeline chaining together dozens of model calls to complete a task.

What’s driving the surge

Cost-efficient open-weight models, particularly those developed by Chinese AI labs, have gained significant traction in high-volume coding and automation use cases.

OpenRouter’s 2025 annual report cited 100 trillion total tokens processed across the platform and flagged the rise of agentic inference as one of its most prominent trends.

The platform’s role as a routing layer gives it an unusually comprehensive view of how models are actually being used across the ecosystem. Developers route traffic through OpenRouter to access multiple model providers through a single API, which means the platform’s data captures usage patterns across a broad cross-section of applications rather than reflecting a single product’s user base.