Insider Brief
- Multiverse Computing said its HyperNova 60B large language model ranked as the most efficient model in the 40-billion to 150-billion parameter category in an independent evaluation conducted by Artificial Analysis.
- According to the company, HyperNova 60B achieved an Intelligence Index score of 29.3 while using 60 billion parameters, making it one of only two models placed in the benchmark’s top category for balancing performance and model size and the only European-developed model in that group.
- Multiverse said the model was built using its CompactifAI optimization technology and is designed to deliver advanced reasoning capabilities with lower computing requirements, potentially reducing deployment costs and infrastructure demands for enterprise users.
Multiverse Computing announced its HyperNova 60B large language model ranked as the most efficient model in the 40-billion to 150-billion parameter category in an independent evaluation conducted by Artificial Analysis.
According to the Spanish artificial intelligence company, HyperNova 60B achieved an Intelligence Index score of 29.3 while using 60 billion parameters, making it the smallest model among those ranked in the top tier for balancing performance and model size.
Artificial Analysis evaluates models by comparing intelligence scores against total parameter counts. HyperNova 60B was one of only two models placed in what the firm describes as the most attractive segment of the comparison, combining relatively high performance with lower computational requirements. It was also the only European-developed model in that group.
Why it Matters
For businesses, smaller models can offer several practical advantages, including lower inference costs, higher throughput on existing hardware and the ability to deploy AI systems in environments where larger models may be impractical because of cost, infrastructure or latency constraints, the company said.
Artificial Analysis’ Intelligence Index combines results from 10 benchmarks covering areas such as reasoning, coding, tool use, long-context understanding, scientific knowledge and instruction following.
Among the reported results, HyperNova 60B scored:
- 29.3 on the Artificial Analysis Intelligence Index
- 67% on GPQA Diamond, a benchmark focused on advanced scientific reasoning
- 58% on IFBench, which measures instruction following
- 63% on τ²-Bench Telecom, which evaluates agentic tool use in telecommunications tasks
- 40% on AA-LCR, a long-context reasoning benchmark
- 33% on SciCode, a scientific coding benchmark
Multiverse said HyperNova 60B was developed using its CompactifAI optimization technology, which is designed to reduce the size of trained models by removing mathematical redundancy while preserving reasoning, instruction-following and tool-use capabilities.
The company describes HyperNova 60B as an open reasoning model intended to provide advanced AI capabilities in a form factor that organizations can deploy more easily than larger frontier models. The model is publicly available through Hugging Face.
Image credit: Multiverse Computing
