Multiverse Computing is introducing Carina 60B, an improved version of HyperNova 60B, our open source 60B model and the most efficient model in its category in the world, validated by Artificial Analysis.
Carina 60B keeps the reasoning, tool calling, and agentic capabilities teams already rely on, and adds a major upgrade: native, high quality Spanish language performance. Carina 60B is available now, both for private deployments and through our CompactifAI API.
Most large language models are trained primarily in English and treat other languages as a secondary capability. Carina 60B closes that gap. It boosts Spanish capabilities alongside full English performance, so teams operating in Spanish speaking markets no longer have to trade language quality for model capability.
Across general intelligence benchmarks, Carina 60B builds on the efficiency profile that already made HyperNova 60B the most efficient model in its category, as externally validated by Artificial Analysis, while closing the gap with larger models on the tasks that matter most for Spanish-language deployments.
Under the hood, Carina 60B uses a Mixture-of-Experts architecture, shaped by CompactifAI, Multiverse Computing's model optimization technology. This keeps Carina 60B efficient enough to run on a single server (20 to 60GB VRAM), while retaining tool calling, general reasoning, and agentic workflows.
Carina 60B is available now, for private deployments and through our CompactifAI API.