New open-source AI model runs without Nvidia

The mystery is finally solved. After six days of Ox Alpha being the most-used model on OpenRouter without anyone knowing its identity, Z.ai has officially revealed it to be their new GLM-5.3-Flash. The announcement came on August 26, 2026, putting an end to the speculations that had animated the AI community in the preceding weeks.

GLM-5.3-Flash represents a significant step forward for open-source models. With 320 billion total parameters (of which 18 billion are active per token), a 1 million token context window, and native capabilities to process text, images, and video, the model positions itself among the most advanced publicly available solutions. The weights are released under an MIT license, allowing for flexible and customized use.

The performance is particularly impressive in coding and agentic task benchmarks. Z.ai reports scores of 84.3 on Terminal-Bench 2.1, 63.4 on DeepSWE v1.1, and 78.4 on Toolathlon Verified. Artificial Analysis assigns the model a score of 57 on its Intelligence Index, confirming its advanced reasoning capabilities.

The most revolutionary aspect might be the price: $0.15 per million input tokens and $0.50 for output, with a 50% discount valid until September 9, 2026. This combination of performance, multimodality, and accessible cost makes GLM-5.3-Flash one of the most compelling value propositions in the current AI landscape.