Z.ai's GLM-5.3-Flash model handles 45% of AI workloads for free

Z.ai announced on August 26 that the mysterious model Ox Alpha, which had been quietly dominating OpenRouter’s traffic, is actually its own GLM‑5.3‑Flash. The company said the model is free to use and can process a large share of AI workloads—up to 45 % of the traffic the platform handles. The announcement came after a week of community forensics that identified the model as a high‑throughput, cost‑effective engine.

OpenRouter now hosts more than 400 models, with roughly ten new ones added every week. Ox Alpha stood out because it was not only free, but also fast and capable of handling several trillion tokens per day. Hobbyists and indie developers began to push the model hard, with estimates of its weekly usage ranging from single digits to over 20 trillion tokens. The community ran tokenizer traces and network analyses to try to determine who could be behind the model and how they could serve so many tokens for no cost. Speculation ranged from a U.S. research lab, to Anthropic’s middle‑tier offering, to Elon Musk’s rumored excess capacity.

Z.ai’s confirmation that GLM‑5.3‑Flash is the model that had been quietly powering OpenRouter’s traffic shows how open‑source and free models can compete with paid services. The model’s architecture is designed for low latency and high throughput, making it attractive to developers who need to run large workloads on a tight budget. Community reaction has been largely positive, with many praising the model’s performance and the transparency of its open‑source nature. The fact that a free model can handle a significant portion of AI traffic challenges the assumption that high‑volume workloads require expensive, proprietary solutions.

The identification of GLM‑5.3‑Flash is likely to shift the competitive landscape. Developers will have a new, powerful option that does not require a subscription, and it may spur further innovation in open‑source AI models. Z.ai has said it will continue to refine the engine and expand its capabilities, while OpenRouter will likely promote the model as a benchmark for free, high‑performance AI. The move underscores the growing importance of community‑driven models and the potential for large‑scale, cost‑effective AI services to shape the future of the industry.