16 local LLMs benchmarked on a Mac mini M6 32GB — top speed ~64 tok/s, raw data released
JeremyCMorgan · x · 2026-10-20
Hans-Dieter Zucht published decode-speed benchmarks for 16 local LLMs on a Mac mini M6 (32GB) in LM Studio, comparing llama.cpp, MLX, and Splash speculative decoding. The fastest model hit roughly 64 tokens per second on prose.
- Runs were done Oct 3–4, 2026; the Mac mini's voltage regulators hit 95°C at minimum fan speed during testing.
- The GitHub repo (MIT license) ships raw data and a benchspeed.py script so others can rerun the benchmark on the same hardware.
- A rare same-hardware comparison for anyone evaluating Apple Silicon local deployment and speculative decoding gains.
More from Infra
- Amazon's $220B AI bet: warehouses of uninstalled GPUs and no 'clear forecasted path' to deploy them — SumitGup · 2026-10-20
- China wasted 360 TWh of wind and solar in half a year — enough to power France for 10 months — ccerrato147 · 2026-10-20
- Bittornado subnet mines root network for recursive self-improvement; 88 Root tops validators — const_reborn · 2026-10-20
- AMD GAIA 0.25 ships with several improvements for local AI agents on Ryzen AI — Fcking_Chuck · 2026-10-20
- SGLang ships Low-Latency Mode with NVIDIA and AMD: up to 1.56× faster P90 token speed — xiaosun86 · 2026-10-20
- PyTorchCon North America 2026 Opens in San Jose With Seven Tracks — PyTorch · 2026-10-20