AetherLabs launches AetherCode: low-latency multimodal model for edge devices

Tech · 4 min read

AetherLabs unveiled AetherCode, a 1.2B-parameter multimodal model optimized for audio, image, and text tasks on ARM and RISC-V hardware. The company claims AetherCode achieves comparable accuracy to larger cloud models on common tasks while operating within strict latency and memory budgets for consumer devices.

Alongside the model, AetherLabs released an SDK for on-device deployment, which includes quantization pipelines, a runtime optimized for heterogeneous compute, and privacy-preserving telemetry. The SDK supports hot-swapping models so manufacturers can update capabilities without full firmware updates.

To accelerate partnerships with device makers, AetherLabs closed a $12 million strategic round led by Vertex Growth and hardware partner Halcyon Partners. Funds will be used to certify AetherCode across vendor silicon and to launch a developer program focused on appliance and automotive integrations.