Spanish ‘soonicorn’ Multiverse Computing releases free compressed AI model

Multiverse Computing, a startup based in Spain, is tackling the challenges posed by large language models. These models, while powerful, often present significant barriers due to their size and associated costs. To mitigate this issue, Multiverse has introduced compressed models designed to make advanced AI capabilities more accessible to businesses. At the heart of this innovation is CompactifAI, a compression technology influenced by quantum computing principles. This technology has been applied to models previously released by OpenAI. As of today, developers can freely access the latest iteration of Multiverse's HyperNova 60B model on Hugging Face. The company has also announced plans to open-source additional compressed models by 2026 to widen their applicability. Multiverse claims that its models are not only smaller but also maintain a high level of potency and accuracy. The HyperNova 60B model, at 32GB, is approximately half the size of its predecessor, OpenAI’s gpt-oss-120B, while offering reduced memory usage and faster response times. The new version, dubbed HyperNova 60B 2602, enhances support for tool calling and agentic coding, particularly in scenarios where inference costs can be substantial. In a competitive landscape, Multiverse asserts that HyperNova 60B outperforms rivals such as Mistral Large 3, developed by the French company Mistral AI. Beyond competition, both companies share similarities in their growth trajectories, as Multiverse has expanded its reach beyond Spain with offices in the United States, Canada, and various European nations, catering to enterprise clients like Iberdrola, Bosch, and the Bank of Canada. Although Multiverse has not yet reached unicorn status, it is reportedly seeking to raise €500 million in funding, with a valuation exceeding €1.5 billion. The company has confirmed that discussions with potential investors are underway but refrained from commenting on the specifics of valuation or funding amounts at this time. Furthermore, Multiverse has not verified claims that its annual recurring revenue (ARR) reached €100 million earlier this year. If accurate, this would still be significantly less than OpenAI's $20 billion ARR but relatively close to Mistral's own revenue, which exceeded $400 million, driven by increasing demand for alternative tech solutions. Positioning itself as a provider of "sovereign solutions across the AI stack," Multiverse recently secured a collaboration with the Aragón regional government in Spain. The company also garnered support from the Basque region, which may soon celebrate the emergence of its first unicorn.