HyperNova 60B ranked first for efficiency in Artificial Analysis' 40-billion-to-150-billion-parameter category, posting a 29.3 Intelligence Index score with 60 billion parameters.
Only two models landed in the benchmark's top efficiency segment, and HyperNova 60B was both the smallest model there and the only European-developed entrant.
Multiverse said its CompactifAI optimization technology cut mathematical redundancy while preserving reasoning, instruction-following and tool-use capabilities, aiming to lower compute needs and deployment costs.
Benchmark details showed 67% on GPQA Diamond, 58% on IFBench, 63% on τ²-Bench Telecom, 40% on AA-LCR and 33% on SciCode; the open model is available on Hugging Face.
For enterprise users, the result highlights a trade-off increasingly prized in AI deployment: smaller models that can deliver strong performance with lower inference costs, higher throughput and less infrastructure.
How does 'quantum-inspired' tech shrink AI models by 95% while preserving their reasoning abilities?
Can Europe's most efficient model challenge the costly dominance of US-based LLMs in business?
HyperNova 60B Sets New Benchmark for Parameter Efficiency: Quantum-Inspired Compression Redefines LLM Performance and Accessibility in 2026
Overview
Multiverse Computing's HyperNova 60B (version 2605) has set a new benchmark in the large language model landscape by becoming the most parameter-efficient model in the 40B–150B open-weights size class. This achievement, confirmed by an independent evaluation from Artificial Analysis, marks a pivotal moment by showing that high intellectual performance can be achieved with far fewer computational resources. HyperNova 60B’s exceptional balance of intelligence and size establishes a new standard for efficiency in LLM development, demonstrating that advanced AI capabilities are possible without the need for massive, resource-intensive models.