Glimmer System One pairs Muse Glimmer 30B with small defect-inspection heads. See its known-product results, image requirements and public code.
By Aryan Srivastava · Public release: 4 Nov 2026
A visual-inspection decision model built around Muse Glimmer 30B. The release supplies two small logistic heads for a fixed manufacturing-defect question, plus a temperature for zero-shot scoring; the backbone is downloaded separately. A head must stay paired with its exact prompt, image resize and object noun.
Facts
Size
30B backbone; each logistic head file is 134,696 bytes
Licence
Heads and code Apache-2.0; MVTec AD data CC BY-NC-SA 4.0, research use only
Inputs
English questions and part images resized to a 448-pixel long side
Access
Public heads and calibration files; Muse Glimmer 30B required separately; code and demo on GitHub
Claims and sources
The maker reports AUROC 0.962 and 90.2% accuracy on the held-out 866-image half of the same 15 MVTec AD products used to train the head. The layer-13 head reports 0.954 and 88.3%. These are known-product tests, not general defect-detection accuracy. vendor reportedsource
The card reports 45 ms per image batched on one H100 for the full model and 21 ms for a truncated layer-13 benchmark. The supplied server still runs the full depth, so loading the early head alone does not give the 21 ms path. vendor reportedsource
Visual-inspection demo and benchmark for Muse Glimmer 30B with logistic decision heads. The maker reports 90.2% defect accuracy on 866 held-out images from known MVTec AD products. Half of each product's test images trained the head. Code, prediction records and a separate zero-shot path are public; the dataset is restricted to research use. The 21 ms early-exit measurement uses a truncated benchmark, while the supplied server runs full depth.