Multiverse Computing pushes its compressed AI models into the mainstream

Multiverse Computing, a Spanish startup, is pushing its quantum-inspired compression technology into the mainstream with two key launches: the CompactifAI chat app and a new self-serve API portal. The company specializes in compressing large models from major labs like OpenAI, Meta, DeepSeek, and Mistral into smaller versions that can run locally on devices. Their chat app features 'Gilda,' a tiny model that operates offline, offering privacy by keeping data on-device. However, it automatically switches to cloud-based models via an API (dubbed 'Ash Nazg') if a device lacks sufficient RAM, which negates the privacy benefit. The app itself has seen limited consumer adoption, with fewer than 5,000 downloads last month.

The strategic focus is clearly on the enterprise. The newly launched API portal gives developers direct access to Multiverse's compressed models, bypassing marketplaces like AWS. CEO Enrique Lizaso highlights features like real-time usage monitoring, giving businesses the transparency and control needed for production deployment. The core value proposition is cost reduction and reduced dependency on external cloud infrastructure, a pressing concern as VC firm Lux Capital warns of financial instability in the AI supply chain. Multiverse's latest compressed model, HyperNova 60B 2602, is based on OpenAI's gpt-oss-120b and claims to deliver faster, cheaper performance than the original, signaling that the capability gap between large and small models is narrowing.