AnalysisAI ModelsAugust 28, 2026

Qwen3.8 27B runs locally on Mac Studio at ~14 tokens/s

Qwen3.8 27B (Q4_K_M, 17GB) generates at ~14 tokens/s on a Mac Studio M3 Ultra via Ollama, while its predecessor qwen3.6:27b does ~28.6 tokens/s. The 1-bit quant (6.7GB) runs at 27 tokens/s in llama.cpp but cannot commit to an answer.

1 source

More stories today

Open the live feed