Spanish Startup Multiverse Computing Releases Free Compressed AI Model Outperforming Mistral

Spanish AI startup Multiverse Computing has made a bold entry into the competitive open-source large language model (LLM) arena with the release of a compressed, free-to-use version of its HyperNova 60B model on Hugging Face. According to the company’s official announcement, the new iteration outperforms Mistral AI’s widely adopted models in key benchmark tests—including reasoning, coding, and multilingual comprehension—while operating at significantly reduced computational cost. The release, which comes without licensing fees or usage restrictions, is being hailed as a potential game-changer for researchers, developers, and organizations seeking high-performance AI without the infrastructure overhead.

Multiverse Computing, often dubbed a ‘soonicorn’ due to its rapid valuation growth and government-backed funding, specializes in quantum-inspired computing solutions for enterprise applications. Founded in 2020 and headquartered in Bilbao, the company has previously partnered with major financial institutions and pharmaceutical firms to optimize complex simulations using quantum algorithms. Its foray into AI model development represents a natural extension of its core mission: leveraging advanced computational paradigms to solve problems beyond classical limits. The HyperNova 60B model, originally trained on a diverse corpus of scientific, technical, and multilingual text, has been distilled into a more compact architecture without sacrificing accuracy, according to internal benchmarks provided by the company.

While the name ‘Multiverse Computing’ may evoke cosmic theories, the firm has repeatedly clarified that its branding is metaphorical, referencing the vast computational possibilities enabled by its hybrid quantum-classical systems—not literal parallel universes. This distinction is critical, as confusion occasionally arises from unrelated scientific articles. For instance, New Scientist has published multiple pieces on theoretical physics research into the multiverse hypothesis, including quantum experiments aimed at detecting evidence of parallel universes. However, these studies, while fascinating, are entirely unrelated to the startup’s AI work. The conflation of terms has led to occasional media misreporting, but Multiverse Computing’s technical documentation and public statements consistently emphasize its focus on computational efficiency and real-world applications.

The release of HyperNova 60B on Hugging Face follows a broader trend of open-weight models challenging proprietary AI systems. Unlike closed models from OpenAI or Google, open models allow for transparency, customization, and community-driven improvement. Multiverse Computing’s decision to make its model freely available suggests a long-term strategy to build developer adoption and ecosystem loyalty, potentially positioning itself as a key player in the next wave of AI infrastructure. Early adopters report that the compressed model runs efficiently on consumer-grade GPUs, making advanced AI accessible to smaller labs and emerging markets.

Industry analysts note that the timing of the release is strategic. With regulatory scrutiny increasing on AI giants and rising costs associated with training massive models, lightweight, high-performance alternatives are gaining traction. ‘Multiverse Computing is betting on efficiency as the new frontier,’ said Dr. Elena Ruiz, an AI policy researcher at the University of Barcelona. ‘They’re not trying to out-train the giants—they’re out-engineering them.’

The company has not disclosed the exact compression techniques used, citing proprietary algorithms, but confirmed that the model retains 98% of the original HyperNova 60B’s performance on the MMLU and GSM8K benchmarks. The release includes full documentation, sample inference code, and a detailed comparison chart against Mistral 7B and Llama 3 8B. Community feedback on Hugging Face has been overwhelmingly positive, with users praising its low latency and accuracy in Spanish and Catalan language tasks—a nod to its European roots.

As the AI race intensifies, Multiverse Computing’s move may catalyze a new wave of ‘compressed intelligence’—a paradigm where model size no longer dictates capability. With venture capital flowing into efficiency-focused AI startups and governments seeking sovereign AI capabilities, this Spanish innovator could be at the vanguard of a more democratized, sustainable AI future.