Multiverse Computing pushes its own AI models

It’s the fault of the companies they’re running on more than 9.2% – high returns over the years – VC firm Lux Capital recently advised companies that rely on AI to achieve their potential confirmed in writing. Due to the instability of the economy through a group of AI products, Lux warned, the handshake agreement is not enough.

But there is another way entirely, which is to stop relying on external devices completely. Small AI models that run directly on the user’s device – no data center, no cloud provisioning, no peer-to-peer risk – are becoming good enough to be considered. And Multiverse Computing raising his hand.

Spanish startups have so far maintained a lower profile than their peers, but as the demand for AI grows, this is changing. After compressing samples from major AI labs including OpenAI, Meta, DeepSeek and Mistral AI, it has launched an entire program that demonstrates the potential of its compressed models and an API portal – a gateway that allows developers to enter and build with those models – which makes them more accessible.

The CompactifAI softwarewhich shares its name with Multiverse’s quantum-inspired compression technology, is an AI chat tool in the vein of ChatGPT or Mistral’s Le Chat. Ask a question, and the model will answer. The difference is that Multiverse included Gilda, a very small model that can run locally and offline, according to the company.

For end users, this is a taste of AI at the edge, with data that does not leave their devices and does not require integration. But there is a caveat: their mobile devices must have enough RAM and storage. If they don’t — and most older iPhones don’t — the app switches back to cloud-based versions via an API. The path between the local and cloud systems is controlled by a machine that Multiverse has named Ash Nazg, whose name will ring a bell for Tolkien fans as it also references the One Ring text in “The Lord of the Rings.” But when the software goes to the cloud, it loses its main secret in the process.

These limitations mean that CompactifAI is not ready for mass adoption, although that may not be the goal. According to data from Sensor Tower, the program had it less than 5,000 downloads in the last month.

The real goal is business. Today, Multiverse is launching a self-serve API portal which gives developers and businesses direct access to its compressed versions – no AWS Marketplace required.

“The CompactifAI API portal (now) gives developers access to compressed models with the transparency and control needed to improve productivity,” CEO Enrique Lizaso said in a statement.

Real-time usage monitoring is one of the main features of the API, and it’s not dangerous. Along with the potential advantages of edge placement, the low cost of computing is one of the main reasons why businesses are considering small models as an alternative to large language models (LLMs).

It also helps to make smaller samples smaller than they used to be. Earlier this week, Mistral updated its small model family with launch of the Mistral Small 4which is said to be designed simultaneously for socializing, urinating, agency work and thinking. A French company too released forgea system that allows businesses to create standard models, including sub-models that they can choose to use and that they are allowed to use.

The latest Multiverse results also show that the gap with LLMs is narrowing. Its latest version, HyperNova 60B 2602it is built on gpt-oss-120b – an OpenAI example whose source code is publicly available. The company claims to provide quick answers at a lower cost than its predecessor, an advantage that is especially needed in agentic systems, where AI completes complex, multi-program tasks.

Designing small models to work on mobile devices while still being useful is a big challenge. Apple Intelligence they avoided this issue by combining the on-device version with the cloud version. Multiverse’s CompactifAI app can also send requests to gpt-oss-120b via the API, but its main purpose is to show that local models like Gilda and its successors have advantages that go beyond cost savings.

For operators in critical environments, a model that can run locally and without connecting to the cloud provides privacy and resilience. But the biggest value lies in the business opportunities it can unlock — for example, putting AI into drones, satellites, and other applications where connectivity can’t be taken for granted.

The company already serves more than 100 clients worldwide including the Bank of Canada, Bosch and Iberdrola, but expanding its customer base could help it unlock more revenue. After raising a $215 million Series B last year, and now Rumors to raise new funds of € 500 million at a cost of more than €1.5 billion.