Spanish AI firm Quasar 438B found to be based on Zhipu GLM-5.2

Dynamic Watch Beating AI News: Shortly after the Spanish AI company Multiverse Computing promoted Quasar 438B as "Europe's strongest AI model," the community uncovered its connection to Zhipu's GLM-5.2. The Artificial Analysis page now directly labels the model as "Quasar 438B (max, based on GLM-5.2)," explicitly stating it is based on GLM-5.2. Multiverse’s own old documentation also contained clues: on August 5, when Quasar 438B launched in beta, the CompactifAI changelog stated its capabilities were "identical" to those of GLM-5.2. The community thus questions whether Quasar is merely a compressed or modified version of GLM-5.2 repackaged under a new name. Multiverse has not disclosed its architecture, training methodology, or clarified the exact relationship between the two models. Although Multiverse’s core business is model compression, its own FAQ still lists Quasar under "Original Models," not "Compressed Models." GLM-5.2 has approximately 753B total parameters with 40B activated during inference and an Artificial Analysis Intelligence Score of 53 on the max setting; Quasar, reduced to 438B parameters, scores only 43, but achieves an output speed of approximately 174 tokens per second—significantly faster than GLM-5.2’s ~68 tokens per second.