Donostia-Sebastián, June 11th - Multiverse Computing’s HyperNova 60B (version 2605) has been independently evaluated by Artificial Analysis with results that place it as the world’s most efficient model in the 40B–150B size category. HyperNova 60B ranks as the most efficient model among the 15 included in the comparison, as the one that best combines intelligence (29.3 on the Intelligence Index) with a compact parameter count (just 60B).
This places it as one of only two models to land in the most attractive quadrant of the analysis, which measures the relationship between intelligence and model size. It is also the model with the lowest parameter count in the category, and the only European-origin model in the ideal quadrant in the comparison independently conducted by Artificial Analysis.
These results reinforce Multiverse Computing’s standing as a global provider of high-performing, efficient AI models. HyperNova 60B demonstrates that frontier-level reasoning capability does not require massive parameter counts, a thesis at the core of Multiverse Computing’s model family and one that this independent evaluation confirms.
The ideal quadrant: high intelligence, low computational cost
Artificial Analysis publishes a scatter plot that maps Intelligence Index score (vertical axis) against total model parameters on a logarithmic scale (horizontal axis). The upper-left quadrant, marked as the “most attractive quadrant”, is where a model delivers the highest intellectual performance at the lowest parameter cost.
HyperNova 60B 2605 is one of the very few models in the entire comparison that appears in that quadrant, and the one with the lowest parameter count. Models with similar or higher Intelligence Index scores need between 120B and 222B parameters to get there. HyperNova 60B reaches a comparable level at 60B.
For organizations deploying models in production, that difference is not marginal. Fewer parameters means lower cost per generated token, greater concurrency on the same hardware, and viable deployment in on-premise environments, regulated infrastructure, or latency-sensitive systems where a 120B model would simply not be an option.
HyperNova 60B is also the only European model in the comparison to reach the most attractive quadrant, a result that positions European AI not just as a sovereign alternative, but as a competitive one.
The Intelligence Index: what the numbers say
The Artificial Analysis Intelligence Index v4.0 aggregates 10 independent evaluations covering agentic reasoning, tool use, coding, long-context handling, knowledge, scientific reasoning, and instruction following. HyperNova 60B scores 29.3, placing it 6th in the 40B–150B total parameters category, ahead of models including Mistral Small 4, gpt-oss-120b (low), K2 Think V2, LongCat Flash Lite, Llama 3.3 70B, and Llama 4 Scout.
Breaking down by individual benchmark:
GPQA Diamond (PhD-level scientific reasoning): 67% – above gpt-oss-120b (low) and Llama 4 Scout, and competitive with larger models.
IFBench (instruction following): 58% – strong result against models with twice the parameter count.
τ²-Bench Telecom (agentic tool use, telecom domain): 63% – reflecting the model’s readiness for complex, multi-step agentic workflows.
AA-LCR (long context reasoning): 40% – long-context capability preserved after optimization.
SciCode (scientific coding): 33%
Terminal-Bench Hard (agentic coding and terminal use): 18%
GDPval-AA (agentic real-world tasks): 16%
Taken together, the results confirm that HyperNova 60B is not a compact model that trades away capability: it is a model that competes with significantly larger architectures on the metrics that matter in production.
What HyperNova 60B is and how it was built
Built in Europe by Multiverse Computing, HyperNova 60B is an open reasoning model, designed to deliver frontier-level intelligence in a footprint that organizations can realistically deploy. The model has 60B parameters, supports advanced reasoning, and is publicly available on Hugging Face as version 2605.
The model is the result of applying CompactifAI, Multiverse Computing’s proprietary technology for optimizing large language models. CompactifAI identifies and removes mathematical redundancy latent in already-trained neural networks, leaving untouched the patterns the model uses to reason, follow instructions, and use tools. The result is not a degraded approximation of a larger model: it is a Multiverse Computing model in which compactness is a design property, not a trade-off.
This approach, which Multiverse Computing applies across its entire model family, is grounded in a clear thesis: reasoning capability should not require massive scale. The Artificial Analysis results independently confirm it.
HyperNova 60B 2605 is available on Hugging Face: MultiverseComputingCAI/Hypernova-60B-2605
About Multiverse Computing
Multiverse Computing is a leader in sovereign and efficient AI. The company develops fast, efficient, and highly specialized AI models that enable organizations to deploy advanced artificial intelligence securely within their own infrastructure, ensuring full control over data, governance, and compliance. Serving sectors where privacy, reliability, and operational efficiency are critical, Multiverse helps enterprises unlock the value of AI while maintaining sovereignty over their most sensitive information. Headquartered in Donostia-San Sebastián, Spain, with offices in the United States, Canada, and across Europe, Multiverse serves more than 100 global customers, including Iberdrola, Bosch, and the Bank of Canada.
For more information, visit www.multiversecomputing.com