Spanish startup Multiverse Computing releases free AI model it says beats Mistral
Spanish startup Multiverse Computing, labeled a "soonicorn" (a startup on its way to unicorn status), has released a new version of its HyperNova 60B language model on Hugging Face. The company says the compressed model outperforms a comparable offering from Mistral. The model is free, which makes it an attractive alternative for developers and researchers. The release highlights growing competition among open AI models coming out of Europe.
AI-processed from TechCrunch; edited by Hamidun News
Europe's race for open artificial intelligence has a new, unexpected entrant. Spanish startup Multiverse Computing, until now known mostly in the narrow circles of quantum computing, has published a free compressed version of its HyperNova 60B language model on Hugging Face. It came with a bold claim: according to the company, its model beats a comparable model from France's Mistral, one of the flagships of European AI, on quality.
Multiverse Computing is based in San Sebastián and is classed in the industry as a "soonicorn", meaning a startup fast approaching a valuation of one billion dollars. The company initially focused on quantum computing for the financial sector and on optimization problems, but in recent years it has actively expanded its expertise in large language models. HyperNova 60B is the result of that strategic pivot, and its release in open access signals serious ambitions.
The approach is technically interesting in itself: this is not just the publication of yet another large model, but the release of a compressed version of one. Compressing language models is one of the hottest areas of modern AI engineering. The goal is to preserve the answer quality of a model with tens of billions of parameters while sharply cutting its computing requirements.
That means the model can run on cheaper hardware, generate answers faster and reach a wider circle of developers. Multiverse Computing apparently uses its own work on quantum-inspired optimization algorithms to achieve a high degree of compression without a critical loss of quality, although the company is so far disclosing details of the methodology only in small doses.
The claim of superiority over Mistral is both brazen and calculated. Mistral AI, founded by former researchers from Google DeepMind and Meta*, has become a symbol of European ambition in generative AI. Its models, notably Mistral Large and the more compact variants, are considered among the best open solutions on the market. Challenging this particular player means announcing yourself at the highest level. Still, such claims deserve caution: without independent benchmarks and a detailed comparison across a variety of tasks, assertions of superiority remain a marketing move. The developer community on Hugging Face will almost certainly run its own tests in the coming weeks, and the picture will then become clearer.
The fact that the model was published for free and in open access deserves attention of its own. In 2026 the AI model market is increasingly splitting into two camps: proprietary solutions from large corporations such as OpenAI and Google, and an ecosystem of open models, where Meta* with Llama, Mistral and now Multiverse Computing compete. For developers, startups and researchers on limited budgets, every new high-quality open model expands what is possible. A compressed model that can run on affordable hardware is especially valuable for small teams that cannot afford clusters of hundreds of GPUs.
For the European AI ecosystem, the emergence of a strong Spanish player alongside France's Mistral and Germany's Aleph Alpha is a positive sign. Europe has long been criticized for lagging behind the US and China in the AI race, with most of its effort going into regulation rather than building technology. Multiverse Computing shows that innovation can also come from outside the usual tech hubs, from Basque San Sebastián and not only San Francisco or Paris.
The main question now is whether HyperNova 60B will live up to its ambitions in practice. If independent tests show that the compressed model is genuinely competitive with Mistral's, it may shift the balance of power among open models and draw the attention of major investors to Multiverse Computing. Either way, the fact that a small European startup with roots in quantum computing is throwing down the gauntlet to established leaders says something about the maturity and dynamism of the market. Competition in open AI is intensifying, and everyone benefits.
*Meta has been designated an extremist organization and banned in Russia.
- Nov 1, 202616 models tested on 7 search tools: Exa averaged 10.5 out of 14, SearXNG about 3
- Oct 30, 2026Mistral opens Large 4 weights: strong at code, but Chinese models score 32 to its 26
- Oct 30, 2026Claude Enterprise Gets Smart Reports: Who Uses AI How, and What It Costs
- Oct 30, 2026SkyRL and vLLM Fine-Tune Mistral Large 4 with RL: AIME Score Rises from 64% to 75%
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.