Mistral unveils Breeze, a lightweight on-device LLM for sketch-to-code conversions

Tech · 4 min read

Breeze is designed to interpret screenshots and vector inputs and generate platform-specific UI scaffolding (React Native, Flutter) while keeping latency low. Mistral’s engineering blog highlights optimizations for quantized execution and model pruning to hit interactive speeds on-device.

The release includes an SDK that integrates with popular design tools, allowing local inference during design sessions and enabling teams to keep IP on-premises. Mistral balanced capability with size: Breeze handles many typical UI elements and interactions but defers highly nuanced visual reasoning to larger cloud models.

Mistral is pursuing partnerships with OEMs to preinstall Breeze-powered tooling on productivity devices and is offering enterprise licensing that bundles model updates with private hosting options. The model’s on-device approach reflects growing demand for offline, privacy-preserving design assistants.