Multiverse Computing built a business around compressing large AI models. Now it has an app. TechCrunch confirmed that the company has compressed models from OpenAI, Meta, DeepSeek, and Mistral AI, and launched both a consumer-facing application and a partnership with cloud infrastructure provider Cerebrium.
The compression technology is called CompactifAI. Multiverse Computing says it uses quantum-inspired techniques, the company describes itself as “a leader in quantum-inspired AI model compression” on its own website. According to Multiverse Computing, CompactifAI can compress models by up to 95% with what the company describes as only 2-3% precision loss. Those figures have not been independently evaluated by Epoch AI or any published arXiv study. Treat them as vendor claims until external benchmarks exist.
The partnership with Cerebrium brings compressed model deployment to cloud infrastructure, per the company’s announcement. That combination, an app for end users, a cloud path for developers, suggests Multiverse Computing is moving from research-oriented compression work into production distribution.
Why this matters to practitioners: smaller models that retain most of their capability are useful in constrained inference environments. Edge deployment, on-device AI, cost-sensitive cloud workloads. The question is whether the claimed performance holds at scale under independent testing. A 95% size reduction with 2-3% quality loss, if it survives scrutiny, changes the math on what hardware is needed to run capable AI. If it doesn’t, the app is a proof-of-concept. Independent evaluation is the next development to watch.