Thunderobot fits a 120B model behind 64GB of RAM

Thunderobot’s AI Master M7000 claims local GPT-OSS-120B inference with 64GB of unified memory plus an 85GB SSD-based aiDAPTIV+ cache—not 149GB of RAM. The TechRadar report says the cache holds less-active model state and mixture-of-experts data when memory gets tight, while Phison’s aiDAPTIV overview describes it as retained capacity rather than immediate compute memory. For local-model builders, the useful number is the 85GB caveat: storage-backed capacity can make a larger model boot on a laptop, but it does not erase latency and bandwidth tradeoffs behind Google’s Android memory crunch.