Every open model in our directory, ranked by how well it runs on Mac mini (M6) (16GB) at 4-bit. Grades weigh whether the model fits in memory and how fast it should run.
Mac mini (M6) has 16 GB of VRAM, enough to run 246 of the 414 open models we track at 4-bit. The largest that still fits well is Llama 2 13B Chat (13B), and the top-rated is Mellum2.1 at about 61 tok/s.
246 of 414 models run comfortably on Mac mini (M6).
See full Mac mini (M6) specs and pricingThe top models this device can run at 4-bit, ranked by fit and speed.
| Model | Grade | Speed | VRAM |
|---|---|---|---|
| Mellum2.1JetBrains | SS | 60.6 tok/s | 7.5 GB |
| LFM2.5-8B-A1BLiquid AI | SS | 42.4 tok/s | 6.8 GB |
| Falcon3-3B-BaseTechnology Innovation Institute (TII) | AA | 50.5 tok/s | 2.4 GB |
| LFM3-Omni-2BLiquid AI | AA | 53.2 tok/s | 2.3 GB |
| MiniCPM-OCR 1.0openbmb | AA | 73.3 tok/s | 1.7 GB |
| ALoDLM-8BAmazon | AA | 22.8 tok/s | 5.4 GB |
| Apertus 8BEPFL, ETH Zurich, CSCS | AA | 22.8 tok/s | 5.4 GB |
| Llama 3 8B InstructMeta | AA | 21.7 tok/s | 5.7 GB |
| Cola2ByteDance-Seed | AA | 21.0 tok/s | 5.9 GB |
| Gemma 4 E2B ITGoogle | AA | 33.2 tok/s | 3.7 GB |
| LensVLM-9BApple | AA | 20.5 tok/s | 6.0 GB |
| Carnice-9b for Hermes agentkai-os | AA | 20.5 tok/s | 6.0 GB |
Mellum2.1JetBrains | 12B(2.5B active) | SS | 60.6 tok/s | 7.5 GB | |
LFM2.5-8B-A1BLiquid AI | 8.3B(1.5B active) | SS | 42.4 tok/s | 6.8 GB | |
Falcon3-3B-BaseTechnology Innovation Institute (TII) | 3B | AA | 50.5 tok/s | 2.4 GB | |
LFM3-Omni-2BLiquid AI | 2.17B | AA | 53.2 tok/s | 2.3 GB | |
MiniCPM-OCR 1.0openbmb | 1.3B | AA | 73.3 tok/s | 1.7 GB | |
ALoDLM-8BAmazon | 8B | AA | 22.8 tok/s | 5.4 GB | |
Apertus 8BEPFL, ETH Zurich, CSCS | 8B | AA | 22.8 tok/s | 5.4 GB | |
| 8B | AA | 21.7 tok/s | 5.7 GB | ||
| Ad | |||||
Cola2ByteDance-Seed | 8.77B | AA | 21.0 tok/s | 5.9 GB | |
Gemma 4 E2B ITGoogle | 2B | AA | 33.2 tok/s | 3.7 GB | |
LensVLM-9BApple | 9B | AA | 20.5 tok/s | 6.0 GB | |
| 9B | AA | 20.5 tok/s | 6.0 GB | ||
VibeThinker-3BWeiboAI | 3B | AA | 32.3 tok/s | 3.8 GB | |
Llama 2 13B ChatMeta | 13B | AA | 14.5 tok/s | 8.5 GB | |
PersonaPlex 7BNVIDIA | 7B | AA | 25.7 tok/s | 4.8 GB | |
Llama 2 7B ChatMeta | 7B | AA | 25.7 tok/s | 4.8 GB | |
| Ad | |||||
MiniMax M3.1 FlashMiniMax | 0B | AA | 246.3 tok/s | 0.5 GB | |
Jais 3MBZUAI and Inception42 | 0B | AA | 246.3 tok/s | 0.5 GB | |
BolmoAi2 | 0B | AA | 246.3 tok/s | 0.5 GB | |
Falcon3-10B-BaseTechnology Innovation Institute (TII) | 10B | BB | 13.8 tok/s | 8.9 GB | |
MiniCPM5-2Bopenbmb | 2.52B | BB | 26.8 tok/s | 4.6 GB | |
MiniCPM-V 4.7 1Bopenbmb | 1.55B | BB | 26.5 tok/s | 4.6 GB | |
Youtu-Parsing-Omnitencent | 5.33B | BB | 19.3 tok/s | 6.4 GB | |
Falcon3-Mamba-7B-BaseTechnology Innovation Institute (TII) | 7B | BB | 19.3 tok/s | 6.4 GB | |
| Ad | |||||
Falcon3-7B-BaseTechnology Innovation Institute (TII) | 7B | BB | 19.3 tok/s | 6.4 GB | |

Model Compatibility
Use the calculator to weigh this device against any other, model by model, with speed and fit scores.
Straight answers to the questions we hear most often.
Can't find what you're looking for? Book a Discovery Call
Mellum2.1 is the highest-rated model that runs on Mac mini (M6) at 4-bit, at about 61 tok/s. In total, Mac mini (M6) can run 246 of the 414 open models we track.
We size each model at 4-bit (Q4_K_M), add the KV cache for its context length and a fixed runtime overhead, then compare that to this device VRAM. Each model is graded on how comfortably it fits and how fast it should run on this card.
It is an estimate of decode speed in tokens per second at 4-bit, based on this device memory bandwidth and each model size. Real speed depends on your runtime, batch size, and context length, so use it to compare models, not as a hard promise.
Lower-bit quantization shrinks a model so more of them fit. These grades already assume 4-bit, which is the common local default. Going below 4-bit can squeeze a larger model in at some quality cost; going above needs more VRAM and may not fit.
Mixture-of-experts models only activate a fraction of their parameters at a time, so their runtime memory is closer to the active size than the headline size. We score fit on the active parameters, which is why some very large models still fit.