Europe's First Model Identity Revealed: Quasar 438B Based on the Spectranet GLM-5.2

OneMillion AI News Flash: Spain-based AI company Multiverse Computing has just promoted Quasar 438B as the "Most Powerful AI Model in Europe," but the community has uncovered its relationship with the Intellispectrum GLM-5.2. Artificial Analysis's current page has directly labeled the model as "Quasar 438B (max, based on GLM-5.2)," clearly stating that it is based on GLM-5.2.

Clues were actually present in Multiverse's own old documents. When Quasar 438B went live in beta on August 5th, CompactifAI's update log clearly stated that its capabilities are "exactly the same" as GLM-5.2. Based on this, the community questioned whether Quasar is simply a compressed or modified version of GLM-5.2 under a new name. Multiverse has not yet disclosed the architecture, training methodology, nor provided an explanation of the specific relationship between the two models. Although Multiverse's main business is model compression, its FAQ currently lists Quasar under "Original Models" without categorizing it under "Compressed Models."

GLM-5.2 has a total of approximately 753 billion parameters, with 40 billion activations during inference. The Artificial Analysis Intelligence Index rates it at 53 points in the max category. After compressing down to 438 billion, Quasar only scores 43 points but achieves an output speed of approximately 174 Tokens/s, significantly faster than GLM-5.2's 68 Tokens/s.