Khazix tests ASUS ProArt 16 RTX Spark laptop for running local AI models

Khazix (数字生命卡兹克) says he spent a week testing the ASUS ProArt 16 with RTX Spark, priced at 37,999 yuan on JD.com, and ran local AI models on it.

He reports a local Qwen 3.8-27B model at about 30 tokens per second, plus image, video, OCR and translation models. The machine has 128GB of unified memory and a 1TB SSD.

He argues local models remove per-token costs, which he says reached 6,000 yuan a month for his AI subscriptions. In a separate passage he says a preset model on the machine runs at 80 tokens per second and that all of these operations run on the local computer at no cost. The 30 and 80 tokens-per-second figures appear in the same report without a stated reason for the difference.