Optimization Lets Qwen3.8-Flash-Next Run on 37GB VRAM

A developer enabled Qwen3.8-Flash-Next to run all experts on just 37GB of memory by streaming 60% of expert weights from disk, reaching 40-73 tok/s on M4 Max. HamsterResearch built on pMLX with dynamic quantization and expert streaming that auto-adapts to task needs.

2026-08-29 ~ 2026-08-30 · 4 related posts

• Qwen3.8-Flash-Next runs with full experts on just 37GB RAM — EyalToledano · 2026-08-29
• Qwen3.8 Optimization: Disk Streaming Yields 40 tok/s — samfajreldines · 2026-08-29
• Qwen3.8 Optimization Hits 73 tok/s via Disk Streaming — EyalToledano · 2026-08-29
• HamsterResearch Enables Full Qwen Model on M4 Max With 37GB RAM via Streaming — EyalToledano · 2026-08-30