OpenRouter: +25,000% Tokens, Uncertain Real-World Usage

Since January 2025, token consumption on OpenRouter has surged by over 25,000%, but this increase does not necessarily reflect greater adoption. OpenAI's Astra dominates revenue, while GPT 5.6 Luna generates the most tokens. Chinese models Kimi, GLM, and DeepSeek are expected to see their spending multiply tenfold in 2026, but starting from a much lower initial level.

Astra Dominates Revenue, Kimi-GLM-DeepSeek Accelerate in 2026

On OpenRouter, OpenAI's Astra ranks first in terms of revenue generated. At the same time, the Chinese models Kimi, GLM, and DeepSeek are experiencing significant growth, with monthly spending projected to increase tenfold in 2026. However, this growth starts from a much lower base than that of the most widely used models.

The Token Surge Does Not Reflect Adoption

The dramatic increase in the number of tokens on OpenRouter primarily indicates an inflation of metrics rather than a proportional rise in usage or commercial value. Since January 2025, weekly token consumption has jumped from 0.5 trillion to 126.2 trillion, representing an increase of over 25,000%. These figures primarily describe computational volumes, without guaranteeing a corresponding rise in the number of users or use cases.

Reasoning and Agents Inflate the Numbers

Reasoning models produce large quantities of "thought" tokens before delivering a response, which inflates the total measured. With non-optimized agentic AI systems, even a slight increase in activity can lead to a significant spike in consumption, as these systems consume tokens at a very high rate. The dominance of a model in token volume does not necessarily mean it has more users; it may simply generate more tokens per request. Recently, OpenAI's GPT 5.6 Luna has emerged as the leading token consumer on OpenRouter. The platform serves as an interface for integrating AI models into products, with tokens representing the unit of computation tracked.