Multiverse Computing cuts NVIDIA's multimodal model to 61B parameters

Multiverse Computing released Comet 60B on October 15th, a compressed version of NVIDIA's Nemotron 3.5 Super VL that roughly halves the model's total parameter count while retaining text, image and video input, with downloadable weights in three formats. The company's X announcement and technical report position Comet 60B as a way to run a large multimodal system with less memory, though the accuracy and throughput comparisons remain company-reported and unverified by third parties.

Multiverse Computing cuts NVIDIA's multimodal model to 61B parameters Comet 60B keeps text, image and video inputs, with downloadable weights in three formats; Multiverse's benchmark results are company-reported. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published Primary source: Multiverse Computing on X https://x.com/MultiverseCompu/status/2110753872150954468 Why it matters Comet gives Multiverse a public test of its compression business on a prominent multimodal model. The parameter and memory reductions are specific; the accuracy and throughput evidence still comes from the vendor, so deployment results will decide whether the smaller checkpoint translates into lower real-world inference costs. Multiverse Computing released Comet 60B on October 15th, a compressed version of NVIDIA's Nemotron 3.5 Super VL https://runtimewire.com/models/nvidia/nemotron-3.5-super-vl:free that cuts the model's total parameter count roughly in half while keeping text, image and video input. The company's X announcement https://x.com/MultiverseCompu/status/2110753872150954468 and technical report https://multiversecomputing.com/papers/comet-60b-the-new-open-source-model-from-multiverse-computing describe the model as a way to run a large multimodal system with less memory; their accuracy and speed comparisons remain…