AnalysisAI ModelsAugust 30, 2026

Qwen3.8-Flash-Next tuned for Macs hits 185-190 tps prefill

A LocalLLaMA user's Mac optimizations for Qwen3.8-Flash-Next reach 185-190 tokens/sec prefill with MTP enabled, making MTP the default choice with no downsides. A further optimization helps when RAM and cache are small.

1 source

More stories today

Open the live feed
Qwen3.8-Flash-Next tuned for Macs hits 185-190 tps prefill