according to the latest statistics from openrouter, a leading platform in the ai industry, the open-source agent framework hermes agent has reached an average daily token consumption of 291 billion, with weekly usage exceeding 1.75 trillion, successfully claiming the top spot worldwide. behind this impressive performance lies xiaomi’s self-developed mimo large model, which serves as the core driving force.
as the preferred inference engine for hermes agent, the mimo model contributed a staggering 1.45 trillion token calls in just the past month, ranking first among all supported models. this figure not only highlights xiaomi’s deep expertise in foundational ai technologies but also signifies that domestically developed large models have earned widespread recognition from global developers in handling complex tasks.
hermes agent has become the fastest-growing open-source agent project worldwide thanks to its unique “self-evolving” capabilities—featuring persistent memory, autonomous learning, and self-optimizing skills—allowing it to continuously evolve as its usage increases. this highly automated task-processing approach places extremely high demands on underlying large models, requiring exceptional instruction-following accuracy, the ability to handle million‑token long contexts, and sustained stability during extended reasoning processes. the mimo series’ outstanding performance in ultra‑intensive task scenarios fully validates its comprehensive real-world capabilities.
to further empower the global developer ecosystem, xiaomi officially launched the “mimo orbit 100t token program,” offering a total of 100 trillion tokens free of charge to ai users worldwide over a 30‑day period. as of the morning of may 12, nearly 80 trillion tokens had already been distributed. with limited slots and a “first‑come, first‑served” policy, remaining quotas are now extremely tight; interested developers should submit their applications through official channels as soon as possible.
