Multiverse Computing Launches New Comet 60B AI Model
The new model leverages advanced compression to reduce GPU memory usage and infrastructure costs for AI workloads.
Updated on Oct. 15, 2026 in Artificial Intelligence
Live Poll
Would your workplace consider switching to smaller, more efficient AI models to reduce operating costs?
Multiverse Computing has released Comet 60B, a multimodal artificial intelligence model designed to support text, image, and video processing. The model aims to lower infrastructure costs by reducing GPU memory requirements for agentic AI applications.
Why it matters
By optimizing memory usage while maintaining performance, this model helps organizations scale agentic AI workloads more affordably. It provides an efficient alternative for companies seeking to implement high-capability models without the burden of massive hardware footprints.
Comet 60B features 60 billion total parameters and 12 billion active parameters, matching the original model's accuracy within 5 percent. It is available in BF16, FP8, and NVFP4 formats and utilizes Group Relative Policy Optimization.
The players
Multiverse Computing
Headquartered in San Sebastian, Spain, this technology company provides AI and quantum computing solutions to over 100 global customers.
NVIDIA
This multinational technology company develops graphics processing units and computing platforms that power many modern large-scale artificial intelligence models.
The details
Multiverse Computing utilized its proprietary CompactifAI framework to conduct tensor-network profiling, compression, and healing to build the model. The architecture is derived from the NVIDIA Nemotron 3.5 Super VL base model, which contains 120 billion total parameters.
Timeline
June 2026: Multiverse Computing launched the Pulsar 16B model.
October 15, 2026: Multiverse Computing released the Comet 60B model.
The Tech Race
This release follows the industry trend of compressing large-scale foundational models like the NVIDIA Nemotron 3.5 Super VL to improve deployment efficiency. By optimizing models for reduced infrastructure footprints, companies are positioning themselves to dominate the burgeoning agentic AI market.
Developers and enterprises can expect lower GPU memory requirements and reduced infrastructure costs when integrating this model into agentic workflows. These optimizations allow for more scalable deployment of multimodal AI capabilities without the need for increased hardware investment.
The takeaway
The move toward model compression signals a shift in AI development where efficiency is becoming as critical as raw parameter size. Organizations can look to these compressed formats to maintain high accuracy while significantly cutting operational expenses.
Further reading
Learn more about the latest innovations in Artificial Intelligence.
Live Poll
Would your workplace consider switching to smaller, more efficient AI models to reduce operating costs?
