Verge tests 125B-parameter Qwen model locally on M5 Ultra Mac Studio

A Verge writer running AI locally loaded Qwen 3.8 Flash Next, a 125-billion-parameter model — parameters are the adjustable values that set capability — sized at about 105GB, via Hermes Agent's setup-time model picker. The test machine was a 256GB unified-memory M5 Ultra Mac Studio, according to an Oct. 11 diary entry.