14MB Foundation Model Runs on Phones, Wearables, Smart Home Gear

Cactus-compute released Needle, a 14MB foundation model for phones, wearables, smart home devices, and robots, gaining 547 stars on GitHub today.

For operators, this compresses the threshold for on-device inference by roughly two orders of magnitude relative to typical small language models. The immediate implication is a lower bill of materials for edge AI: less flash storage, less RAM, and fewer heat constraints. Builders no longer need to reserve a cloud round-trip for basic classification or generation tasks; a microcontroller-class chip can now host a model without a dedicated accelerator. The second-order effect is architectural: sensor fusion logic can move from the gateway to the endpoint. This obsoletes the current pattern of streaming raw telemetry to a hub for processing. Deployments in battery-constrained environments—agriculture, logistics wearables, smart locks—can now run inference continuously rather than in wake-on-demand bursts. Plan for smaller power budgets and updated over-the-air update pipelines to handle a 14MB binary, but expect latency and privacy metrics to improve proportionally.