The companies said the model uses about 3 GB of active memory.
The October 15 press release claims generation speeds above 13 tokens per second on an iPhone 17 Pro Max and about 17 tokens per second on an iPad Pro M4. It also says the release has day-one chip support. The announcement describes this as a 35B multimodal MoE model, while the A3B suffix indicates the named configuration.
The source is a company press release. It does not provide an independent benchmark, measurement configuration, model download location, licensing terms, or a complete account of device compatibility. It therefore does not establish that the stated memory use or throughput will be reproduced in other applications or on other hardware.