Multiverse Computing unveils free compressed AI model as Spanish soonicorn expands AI reach

Large language models face a basic challenge: they are huge. Multiverse Computing, a Spanish startup, says it is tackling that issue by releasing compressed AI models meant to narrow the gap between what cutting-edge “frontier” systems can do and what organisations can realistically afford to run and deploy.

The company’s key technology is CompactifAI, a compression approach inspired by quantum computing that the Basque startup has applied to models released by OpenAI. Starting today, developers can access a newer version of Multiverse’s HyperNova 60B model for free through Hugging Face. Multiverse also said it intends to open-source additional compressed models in 2026 to support more use cases and broader adoption.

Multiverse argues its compressed models maintain strong performance while being smaller and cheaper to operate. At 32GB, HyperNova 60B is about half the size of the model it is derived from — OpenAI’s gpt-oss-120B — while also offering lower memory requirements and lower latency. The newly released update, HyperNova 60B 2602, adds improved support for tool calling and agentic coding, two areas where inference expenses can quickly rise.

Multiverse has also claimed that HyperNova 60B outperforms Mistral Large 3, a major model released by French AI decacorn Mistral AI. But beyond that competitive angle, the two European companies share several similarities.

Like Mistral, Multiverse has expanded internationally, opening offices in the United States, Canada, and across Europe. Both companies have also built enterprise client bases. Multiverse has listed customers including Iberdrola, Bosch, and the Bank of Canada.

Multiverse is not officially a unicorn, but it is now rumoured to be raising a fresh €500 million funding round at a valuation above €1.5 billion. In a statement, the company said it is in active discussions with potential investors about a new round, but added that it would be premature to comment on the valuation or the size of any funding at this point. Multiverse also declined to comment on reports claiming its annual recurring revenue (ARR) reached €100 million in January.

If that ARR figure were confirmed, it would still represent only a small fraction of OpenAI’s estimated $20 billion ARR. Still, it would be closer to Mistral’s reported trajectory, in which ARR climbed past $400 million, in part due to rising interest in alternatives to U.S.-based technology providers. Multiverse similarly framed its positioning in a recent press release as being able to “deliver sovereign solutions across the AI stack.”

These geopolitical dynamics have also played a role in Multiverse securing partnerships in the home market. The company recently won a collaboration with the regional government of Aragón in northeastern Spain. Spain’s Agency for Technological Transformation (SETT) also participated in Multiverse’s $215 million Series B round last year. Since launching, Multiverse has additionally benefited from backing in the Basque region, which could soon be home to its first unicorn.