Face Intelligence released MiniCPM-V 4.7, a new edge multimodal LLM available in 35B-A3B and 1B variants. The 35B-A3B is the series' first MoE model, powered by LES inference acceleration, reaching 13+ tokens/s on iPhone 17 Pro Max with only 3GB memory. It supports iOS, Android, HarmonyOS and multiple chip platforms, with the SRFT fine-tuning scheme open-sourced.
At the Jamf Nation User Conference (JNUC), Apple shared a chart comparing on-device AI inference capabilities across its hardware lineup, from iPhone and iPad to Mac Studio clusters, showing support ranging from 14 billion active parameters up to 1.6 trillion, powered by unified memory architecture.