AnalysisDevelopersNovember 18, 2026

ToRc: Rust engine runs Qwen3.8 Flash Next on two £300 Snapdragon boxes

Read original source →reddit.com

A from-scratch Rust inference engine runs a 107 GB Qwen3.8 Flash Next model at Q4 across two ~£300 Snapdragon X boxes with 64 GB RAM combined, hitting 67 tok/s prompt and 17 tok/s generation on a 16k prompt. It splits work across CPU, NPU and iGPU and ships as a 13 MB zip.

1 source

More stories today

Open the live feed