Europe Startup Unveils Tiny AI Models for On‑Device Use

Breaking: European AI startup Multiverse Computing rolled out two of the smallest high‑performing models yet, nicknamed SuperFly and ChickBrain, promising offline chat, speech, and reasoning on everyday devices.

What Multiverse built

Multiverse used a quantum‑inspired compression method called CompactifAI to shrink existing open models without losing performance. The startup, founded by quantum computing experts and backed by a recent €189M round, packaged compressed models aimed specifically at on‑device use.

SuperFly is a 94M‑parameter compressed version of SmolLM2‑135, intended for voice interfaces and constrained device tasks. ChickBrain compresses Meta's Llama 3.1 8B down to 3.2B and adds reasoning capabilities while slightly improving several benchmark scores in Multiverse's tests.

Why this matters

Compressing models to run locally changes the economics and privacy profile of AI: lower latency, reduced cloud costs, and data staying on the device. That enables conversational interfaces in appliances, offline assistants on phones and watches, and resilient deployments where network access is limited.

• Voice control for home appliances with minimal hardware (e.g., Arduino‑class controllers)
• On‑device troubleshooting or context‑aware assistants that don’t send user data to the cloud
• Edge reasoning on laptops and phones for offline productivity or regulated environments

Real‑world signals and partnerships

Multiverse says it is already in talks with major device makers including Apple, Samsung, Sony, and HP. It also offers an AWS‑hosted API for developers who want compressed models without deep integration. Customers across industries — from manufacturing to finance — are reported users of its compression for other ML tasks.

What organizations should consider

Compressed on‑device models are not a drop‑in substitute for large cloud models. Evaluate them where low latency, privacy, offline operation, or cost per inference matter most. Consider the following steps to assess fit:

• Benchmark compressed models against your real inputs and edge hardware
• Prototype low‑risk features (voice control, offline assistants) to measure user value and cost savings
• Plan for security, model updates, and regulatory compliance when models run on devices

Bottom line

Multiverse’s SuperFly and ChickBrain showcase how model compression can unlock practical, offline AI across billions of devices. For product teams and policy makers, the arrival of reliable on‑device reasoning and conversational AI means new user experiences and operational tradeoffs. Thoughtful piloting, benchmarking, and integration design will determine winners.

QuarkyByte can help organizations identify the most valuable edge AI use cases, run technical pilots to compare compressed models against cloud alternatives, and design deployment strategies that balance performance, privacy, and cost.