Ox Alpha: Z.AI's GLM Model Shakes Up AI Benchmarks

The mysterious Ox Alpha model has been identified as a new iteration of the GLM series from Chinese lab Z.AI. This model spent a week confusing developers and benchmark trackers. What's interesting is that Ox Alpha appeared on OpenRouter and OpenCode with no company logo. It was described as a reasoning model for coding, sustained agentic work, and production workloads. It also had a context window of over a million tokens, accepting text, image, and video input. The model was offered free through both platforms, with OpenCode advertising a capacity of 100 trillion tokens a day. Within a day of its release, Ox Alpha achieved 80% first-pass accuracy in a 10-task run on the DeepSWE benchmark. This placed it ahead of Claude Fable 5 at 65% and GPT-5.6 Sol at 52%. This single data point made the model go viral. Z.AI plans to release the model's weights tonight, making it open for anyone to use. This development changes the landscape for companies pricing frontier access to AI models.