Open Source Factory

MODELS & COSTS Modified GLM-5.3-Flash: Refusal Scores and Their Limits Open Source Factory · Aug 30, 2026 · 3 min read A lower refusal rate describes behavior on a particular test set. It does not measure every capability or make a modified checkpoint interchangeable with the original. The GLM modifications below should be compared through the teams' stated methods, serving conditions and limits on independent evidence. In this article Refusal changes reported in the weights Reported capability and serving costs How this release fits earlier modifications Linked source documents The numbers OrcaRouter published are the hook. Refusal behavior collapses from 96%, 93%, 97%, and 93% down to 11%, 12%, 15%, and 18% across the four categories they measured, and XSTest — the rate of wrongly refusing harmless requests — drops from 2.4% to 0.4%. Their pitch: a stock GLM-5.3-Flash was over-refusing a lot of benign-but-flagged requests, copyright questions being the clearest real case. ...

MODELS & COSTS Ox Alpha Identified: What GLM's Model Card Added Open Source Factory · Aug 26, 2026 · 4 min read An anonymous endpoint offers little to inspect beyond its behavior. Z.ai's identification of Ox Alpha as GLM-5.3-Flash adds architecture, weights and license terms to that record. Its account of the training hardware remains a developer claim, separate from the fact that the model was identified. In this article Z.ai identifies the anonymous model What the training-hardware claim establishes Reading Z.ai's benchmark table Weights and the MIT license Linked source documents The earlier Ox Alpha article covered the anonymous OpenCode release. The model card now supplies details that the anonymous endpoint could not: architecture, license and the stated training hardware. It still does not provide a controlled energy-efficiency comparison with an Nvidia training run. Z.ai identifies the anonymous model Item GLM-5.3-Flash (Ox Alpha) Parameters 320B total,...