Multiverse Computing Launches Pulsar 16B Reasoning Model Powered by NVIDIA (NVDA) Architecture

Multiverse Computing launched Pulsar 16B, an open reasoning model built with NVIDIA’s Nemotron architecture and optimized using its CompactifAI technology. The model claims 30B-class performance with only 16B parameters, and just 3.1B active parameters, aiming to reduce compute, memory, and deployment costs for enterprise AI use cases. Benchmarks cited in the article suggest it matches or exceeds prior larger models on AIME and GPQA-Diamond, while testing on NVIDIA Blackwell GPUs showed a 43% increase in system throughput and faster time-to-first-token versus the base model. The release, available on Hugging Face under Apache 2.0, reinforces NVIDIA’s role in AI model infrastructure and could support demand for its GPUs and software ecosystem as enterprises pursue scalable, on-premise AI deployments.