Ollama (MLX backend) on MacBook Pro M3 Max 128GB: Qwen3.8-27B 4-bit decode tokens/s
42.5 tok/s (batch 1, short prompt decode)
Fields
id
benchmarks/ollama-m3-max-qwen3-8-27b-mlx-decode-tok-sMetric
decode_tok_s
Value
42.5
Unit
tok/s (batch 1, short prompt decode)
Model
Qwen3.8 27B
Methodology
MacBook Pro M3 Max (16-core CPU / 40-core GPU, 128 GB unified memory at 400 GB/s), macOS 26.6.2. Ollama running with MLX engine backend, NVFP4 format, batch size 1, temp 0, reasoning off, repeated 4x with variance under 2%.
Measured by
third_party
Reproducible
yes
notes
Direct head-to-head empirical measurement on M3 Max 128GB (400 GB/s). Under Ollama MLX backend, Qwen3.8-27B achieves 42.5 tok/s decode throughput, compared to 12.0 tok/s under llama.cpp on the identical machine.
Notes
Direct head-to-head empirical measurement on M3 Max 128GB (400 GB/s). Under Ollama MLX backend, Qwen3.8-27B achieves 42.5 tok/s decode throughput, compared to 12.0 tok/s under llama.cpp on the identical machine.