| Hardware profile | ||||
|---|---|---|---|---|
| Hardware | Memory | Bandwidth | Computing power | vLLM |
| Mac mini M6Apple Mac mini M6 system with 24 GB unified memory | 24 GBUnified RAM | 170 GB/s | ~76 TOPS | no |
| Compatible models Estimated speeds, not benchmark results: calculated from memory bandwidth and model size. Real results can differ significantly because there is no precise formula for deriving LLM generation speed from hardware specifications alone. | |||
|---|---|---|---|
| Model | Size | Approx. Q4 memory | Estimated generation |
| Ornith 1.0 35BMoE | 35Bactive 3B | 21 GB | 37 tok/s4-bit quantization |
| Ornith 1.5 35BMoE | 35Bactive 3B | 21 GB | 37 tok/s4-bit quantization |
| Qwen 3.6 35BMoE | 35Bactive 3B | 21 GB | 37 tok/s4-bit quantization |
| Laguna XS 2.1MoE | 33Bactive 3B | 19.8 GB | 37 tok/s4-bit quantization |
| Qwen 2.5 32BDense | 32.5B | 19.5 GB | 4 tok/s4-bit quantization |
| Gemma 4 31BDense | 31B | 18.6 GB | 5 tok/s4-bit quantization |
| Qwen 3 Coder 30BMoE | 30.5Bactive 3.3B | 18.3 GB | 34 tok/s4-bit quantization |
| Granite 4.1 30BDense | 30B | 18 GB | 5 tok/s4-bit quantization |
| Granite 4.2 30BDense | 30B | 18 GB | 5 tok/s4-bit quantization |
| Muse GlimmerDense | 30B | 18 GB | 5 tok/s4-bit quantization |
| GLM 4.7 FlashMoE | 30Bactive 3B | 18 GB | 37 tok/s4-bit quantization |
| Nemotron 3.5 LightningMoE | 30Bactive 3B | 18 GB | 37 tok/s4-bit quantization |
| Fara 1.5 27BDense | 27B | 16.2 GB | 5 tok/s4-bit quantization |
| Qwen 3.6 27BDense | 27B | 16.2 GB | 5 tok/s4-bit quantization |
| Qwen 3.8 27BDense | 27B | 16.2 GB | 5 tok/s4-bit quantization |
| Gemma 4 26BMoE | 26Bactive 4B | 15.6 GB | 28 tok/s4-bit quantization |
| LFM 2.5 24BMoE | 24Bactive 2B | 14.4 GB | 54 tok/s4-bit quantization |
| GPT-OSS 20BMoE | 20.9Bactive 3.6B | 12.5 GB | 31 tok/s4-bit quantization |
| Instella-MoE 16B ThinkMoE | 16Bactive 2.8B | 9.6 GB | 39 tok/s4-bit quantization |
| Qwen 3 14BDense | 14B | 8.4 GB | 11 tok/s4-bit quantization |
| Gemma 3 12BDense | 12B | 7.2 GB | 12 tok/s4-bit quantization |
| Gemma 4 12BDense | 12B | 7.2 GB | 12 tok/s4-bit quantization |
| Fara 1.5 9BDense | 9B | 5.4 GB | 17 tok/s4-bit quantization |
| Ornith 1.0 9BDense | 9B | 5.4 GB | 17 tok/s4-bit quantization |
| Ornith 1.5 9BDense | 9B | 5.4 GB | 17 tok/s4-bit quantization |
| Qwen 3.5 9BDense | 9B | 5.4 GB | 17 tok/s4-bit quantization |
| Qwythos 9BDense | 9B | 5.4 GB | 17 tok/s4-bit quantization |
| Granite 4.1 8BDense | 8B | 4.8 GB | 19 tok/s4-bit quantization |
| Granite 4.2 8BDense | 8B | 4.8 GB | 19 tok/s4-bit quantization |
| Qwen 3 8BDense | 8B | 4.8 GB | 19 tok/s4-bit quantization |
| LFM 2.5 8BMoE | 8Bactive 1B | 4.8 GB | 100 tok/s4-bit quantization |
| Ling 3.0 TinyMoE | 7.9Bactive 1.3B | 4.7 GB | 80 tok/s4-bit quantization |
| Fara 1.5 4BDense | 4B | 2.4 GB | 37 tok/s4-bit quantization |
| Gemma 3 4BDense | 4B | 2.4 GB | 37 tok/s4-bit quantization |
| Qwen 3 4B Instruct 2507Dense | 4B | 2.4 GB | 37 tok/s4-bit quantization |
| Qwen 3.5 4BDense | 4B | 2.4 GB | 37 tok/s4-bit quantization |
| Spark X2.5 4BDense | 4B | 2.4 GB | 37 tok/s4-bit quantization |
| Granite 4.1 3BDense | 3B | 1.8 GB | 49 tok/s4-bit quantization |
| Granite 4.2 3BDense | 3B | 1.8 GB | 49 tok/s4-bit quantization |
| Llama 3.2 3B InstructDense | 3B | 1.8 GB | 49 tok/s4-bit quantization |
| Ministral 3 3BDense | 3B | 1.8 GB | 49 tok/s4-bit quantization |
| Nanbeige 4.2 3BDense | 3B | 1.8 GB | 49 tok/s4-bit quantization |
| LFM 3 1.7BDense | 1.7B | 1 GB | 85 tok/s4-bit quantization |
Demonstrates local execution of Liquid AI LFM3-1.7B on an Apple Mac mini M6 with 32GB unified memory via OpenClaw, achieving approximately 70 tokens per second even at high context utilization.
Tests Spark-X2.6-4B performance in coding, reasoning, tool calling, and agentic AI workflows on Apple Mac mini M6 with 32GB unified memory via LM Studio and Pi Coding agent.
Tests Strata 0.2 with Qwen3.8-Flash-Next via Apple Metal, covering setup, model loading, memory needs, and prompt processing plus generation speed.
Tests Liquid AI LFM2.5-24B-A2B MoE model on Mac mini M6 via LM Studio and OpenClaw with 128K context, noting activation of ~2.45B parameters per token.
Tests Mirai Uzu against MLX, oMLX, llama.cpp, and MTPLX on M5 Max and M6 Mac mini using speculative decoding and Metal GPU kernels.
Tests Mac mini M6 behavior under concurrent loads of 8 Camtasia 4K exports, local RAG via Ollama, Gemma 4 inference and 4K playback across CPU, GPU, memory, media engines and storage.
Tests local AI on a base M6 Mac mini using Ollama, comparing Gemma 4 26B against Gemma 4 E4B MLX for responsiveness during a setup speed run.
Tests Gemma4-26B MoE, QWEN3.6-35B MoE, Bonsai 2-27B MoE, QWEN3.8-27B and QWEN3.8-27B-Splash via oMLX, MTPLX and LM Studio on Mac mini M6.
Compares Mac mini M6 and M5 Max running Qwen3.8-27B-Splash in LM Studio across sustained generation, TTFT, prefill, multi-turn context, and agentic coding via .NET migration test.