What AI models can a Mac with 16 GB run?
- A Mac with 16 GB fits 13 models for short chats, 10 for long 32,000-token work, from 33 models in our list. We use measured minimums where available and our cautious rule for the rest.
- For long work, the roomiest fit is Qwen3.5-4B. It has the lowest listed memory floor, 6 GB, leaving 10 GB above that floor.
- For long work, the smallest fit by parameter count is Qwen2.5-0.5B-Instruct, listed as 0.5B dense. It takes 0.28 GB on disk. The largest fit by parameter count is Gemma-4-26B-a4b-it, listed as 26B MoE / 4B active.
Every model that fits
Read each pair of minimums before choosing a download. The fit label and spare memory describe long conversations. A tested model keeps its measured result for either chat length.
This tier widens the choice beyond the lowest-memory page. Several models arrive together at the same listed floor, even though their parameter counts differ. That makes the tight label worth reading closely. It means the configured memory equals the source minimum. The room shown in the answer box belongs to the model named there, while each other candidate has its own balance in the table.
Look down the disk column when planning which files to keep. The fitting set includes models whose weight files occupy different amounts of storage. Moving between them changes the download footprint even when the memory requirement stays level. Select a task in the calculator to narrow the list before treating parameter count as a reason to choose a model.
Hardware at this capacity spans several chip families. The same fit decision applies to all of their rows because the memory test uses the configured capacity. The speed cells then give the chip-specific result and its evidence. An older desktop and a newer laptop may share this memory size, so match the full Mac name before reading a rate. The excluded list gives a separate upgrade target for every model still outside this tier.
Compared with the 8 GB memory page, this size adds 6 models. The added models are Qwen3-0.6B, Qwen3-4B-Instruct-2507, gemma-3-4b-it-qat, Qwen2.5-Coder-7B-Instruct, Qwen3.5-9B, Gemma-4-26B-a4b-it. Models already below the earlier limit gain more space above their minimum. The chip changes the speed shown below, while the fit list follows the same memory test.
| Model | Fit | Short chats | Long 32,000-token work | Memory left for long work | Disk |
|---|---|---|---|---|---|
| Qwen2.5-0.5B-Instruct | fits | 8 GB | 8 GB | 8 GB | 0.28 GB |
| Qwen3-0.6B | tight | 8 GB | 16 GB | 0 GB | 0.34 GB |
| Qwen3.5-0.8B | fits | 8 GB | 8 GB | 8 GB | 0.63 GB |
| Llama-3.2-1B-Instruct | fits | 8 GB | 8 GB | 8 GB | 0.7 GB |
| Qwen3-4B-Instruct-2507 | tight | 16 GB | 16 GB | 0 GB | 2.26 GB |
| Qwen3.5-4B | fits | 6 GB measured | 6 GB measured | 10 GB | 2.37 GB |
| gemma-3-4b-it-qat | tight | 16 GB | 16 GB | 0 GB | 3 GB |
| Qwen2.5-Coder-7B-Instruct | tight | 16 GB | 16 GB | 0 GB | 4.28 GB |
| Qwen3.5-9B | fits | 12 GB measured | 12 GB measured | 4 GB | 5.04 GB |
| Gemma-4-26B-a4b-it | tight | 16 GB measured | 16 GB measured | 0 GB | 15.61 GB |
Speed table by chip
The table covers all 22 Mac models sold with 16 GB in our Mac list, including older models. Each row names the exact Mac and chip. The representative models are Qwen3.5-4B, Qwen2.5-0.5B-Instruct, Gemma-4-26B-a4b-it, Qwen3-0.6B. They cover the roomiest fit, the smallest model, the largest model and the middle of this fit list, with repeated names removed.
| Mac and chip | Qwen3.5-4B | Qwen2.5-0.5B-Instruct | Gemma-4-26B-a4b-it | Qwen3-0.6B |
|---|---|---|---|---|
| MacBook Air M1Apple M1 |
about 4.9 to 14 tokens/s
estimate, not measured by us
How we estimate speed4.89 to 13.56 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 68.25 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 49 to 98 tokens/s
estimate, not measured by us
How we estimate speed48.75 to 97.5 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
about 3.0 to 8.2 tokens/s
estimate, not measured by us
How we estimate speed2.97 to 8.25 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 68.25 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 40 to 80 tokens/s
estimate, not measured by us
How we estimate speed40.14 to 80.3 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
| MacBook Pro 13 M1Apple M1 |
about 4.9 to 14 tokens/s
estimate, not measured by us
How we estimate speed4.89 to 13.56 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 68.25 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 49 to 98 tokens/s
estimate, not measured by us
How we estimate speed48.75 to 97.5 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
about 3.0 to 8.2 tokens/s
estimate, not measured by us
How we estimate speed2.97 to 8.25 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 68.25 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 40 to 80 tokens/s
estimate, not measured by us
How we estimate speed40.14 to 80.3 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
| Mac mini M1Apple M1 |
about 4.9 to 14 tokens/s
estimate, not measured by us
How we estimate speed4.89 to 13.56 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 68.25 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 49 to 98 tokens/s
estimate, not measured by us
How we estimate speed48.75 to 97.5 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
about 3.0 to 8.2 tokens/s
estimate, not measured by us
How we estimate speed2.97 to 8.25 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 68.25 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 40 to 80 tokens/s
estimate, not measured by us
How we estimate speed40.14 to 80.3 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
| iMac M1Apple M1 |
about 4.9 to 14 tokens/s
estimate, not measured by us
How we estimate speed4.89 to 13.56 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 68.25 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 49 to 98 tokens/s
estimate, not measured by us
How we estimate speed48.75 to 97.5 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
about 3.0 to 8.2 tokens/s
estimate, not measured by us
How we estimate speed2.97 to 8.25 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 68.25 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 40 to 80 tokens/s
estimate, not measured by us
How we estimate speed40.14 to 80.3 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 68.25 GB/s divided by active weights in GB |
| MacBook Pro M1 ProApple M1 Pro |
about 14 to 35 tokens/s
estimate, not measured by us
How we estimate speed14.34 to 34.85 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 200 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 143 to 286 tokens/s
estimate, not measured by us
How we estimate speed142.85 to 285.72 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 200 GB/s divided by active weights in GB |
about 8.7 to 21 tokens/s
estimate, not measured by us
How we estimate speed8.72 to 21.19 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 200 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 118 to 235 tokens/s
estimate, not measured by us
How we estimate speed117.64 to 235.3 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 200 GB/s divided by active weights in GB |
| MacBook Air M2Apple M2 |
about 7.2 to 21 tokens/s
estimate, not measured by us
How we estimate speed7.17 to 20.94 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 71 to 143 tokens/s
estimate, not measured by us
How we estimate speed71.42 to 142.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
about 4.4 to 13 tokens/s
estimate, not measured by us
How we estimate speed4.36 to 12.74 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 59 to 118 tokens/s
estimate, not measured by us
How we estimate speed58.82 to 117.65 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| MacBook Pro 13 M2Apple M2 |
about 7.2 to 21 tokens/s
estimate, not measured by us
How we estimate speed7.17 to 20.94 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 71 to 143 tokens/s
estimate, not measured by us
How we estimate speed71.42 to 142.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
about 4.4 to 13 tokens/s
estimate, not measured by us
How we estimate speed4.36 to 12.74 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 59 to 118 tokens/s
estimate, not measured by us
How we estimate speed58.82 to 117.65 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| Mac mini M2Apple M2 |
about 7.2 to 21 tokens/s
estimate, not measured by us
How we estimate speed7.17 to 20.94 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 71 to 143 tokens/s
estimate, not measured by us
How we estimate speed71.42 to 142.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
about 4.4 to 13 tokens/s
estimate, not measured by us
How we estimate speed4.36 to 12.74 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 59 to 118 tokens/s
estimate, not measured by us
How we estimate speed58.82 to 117.65 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| MacBook Pro M2 ProApple M2 Pro |
about 14 to 37 tokens/s
estimate, not measured by us
How we estimate speed14.34 to 37.22 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 200 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 143 to 286 tokens/s
estimate, not measured by us
How we estimate speed142.85 to 285.72 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 200 GB/s divided by active weights in GB |
about 8.7 to 23 tokens/s
estimate, not measured by us
How we estimate speed8.72 to 22.63 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 200 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 118 to 235 tokens/s
estimate, not measured by us
How we estimate speed117.64 to 235.3 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 200 GB/s divided by active weights in GB |
| MacBook Air 13 M3Apple M3 |
about 7.2 to 20 tokens/s
estimate, not measured by us
How we estimate speed7.17 to 20.44 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 71 to 143 tokens/s
estimate, not measured by us
How we estimate speed71.42 to 142.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
about 4.4 to 12 tokens/s
estimate, not measured by us
How we estimate speed4.36 to 12.43 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 59 to 118 tokens/s
estimate, not measured by us
How we estimate speed58.82 to 117.65 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| MacBook Air 15 M3Apple M3 |
about 7.2 to 20 tokens/s
estimate, not measured by us
How we estimate speed7.17 to 20.44 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 71 to 143 tokens/s
estimate, not measured by us
How we estimate speed71.42 to 142.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
about 4.4 to 12 tokens/s
estimate, not measured by us
How we estimate speed4.36 to 12.43 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 59 to 118 tokens/s
estimate, not measured by us
How we estimate speed58.82 to 117.65 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| iMac M3Apple M3 |
about 7.2 to 20 tokens/s
estimate, not measured by us
How we estimate speed7.17 to 20.44 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 71 to 143 tokens/s
estimate, not measured by us
How we estimate speed71.42 to 142.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
about 4.4 to 12 tokens/s
estimate, not measured by us
How we estimate speed4.36 to 12.43 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 100 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 59 to 118 tokens/s
estimate, not measured by us
How we estimate speed58.82 to 117.65 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 100 GB/s divided by active weights in GB |
| MacBook Pro 14 M4Apple M4 |
about 8.6 to 23 tokens/s
estimate, not measured by us
How we estimate speed8.6 to 23.09 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 86 to 171 tokens/s
estimate, not measured by us
How we estimate speed85.71 to 171.43 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
about 5.2 to 14 tokens/s
estimate, not measured by us
How we estimate speed5.23 to 14.04 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 71 to 141 tokens/s
estimate, not measured by us
How we estimate speed70.58 to 141.18 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| iMac M4Apple M4 |
about 8.6 to 23 tokens/s
estimate, not measured by us
How we estimate speed8.6 to 23.09 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 86 to 171 tokens/s
estimate, not measured by us
How we estimate speed85.71 to 171.43 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
about 5.2 to 14 tokens/s
estimate, not measured by us
How we estimate speed5.23 to 14.04 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 71 to 141 tokens/s
estimate, not measured by us
How we estimate speed70.58 to 141.18 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| Mac mini M4Apple M4 |
about 8.6 to 23 tokens/s
estimate, not measured by us
How we estimate speed8.6 to 23.09 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 86 to 171 tokens/s
estimate, not measured by us
How we estimate speed85.71 to 171.43 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
about 5.2 to 14 tokens/s
estimate, not measured by us
How we estimate speed5.23 to 14.04 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 71 to 141 tokens/s
estimate, not measured by us
How we estimate speed70.58 to 141.18 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| MacBook Air 13 M4Apple M4 |
about 8.6 to 23 tokens/s
estimate, not measured by us
How we estimate speed8.6 to 23.09 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 86 to 171 tokens/s
estimate, not measured by us
How we estimate speed85.71 to 171.43 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
about 5.2 to 14 tokens/s
estimate, not measured by us
How we estimate speed5.23 to 14.04 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 71 to 141 tokens/s
estimate, not measured by us
How we estimate speed70.58 to 141.18 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| MacBook Air 15 M4Apple M4 |
about 8.6 to 23 tokens/s
estimate, not measured by us
How we estimate speed8.6 to 23.09 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 86 to 171 tokens/s
estimate, not measured by us
How we estimate speed85.71 to 171.43 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
about 5.2 to 14 tokens/s
estimate, not measured by us
How we estimate speed5.23 to 14.04 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 120 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 71 to 141 tokens/s
estimate, not measured by us
How we estimate speed70.58 to 141.18 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 120 GB/s divided by active weights in GB |
| MacBook Air 13 M5Apple M5 |
about 11 to 23 tokens/s
estimate, not measured by us
How we estimate speed10.97 to 22.95 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 109 to 219 tokens/s
estimate, not measured by us
How we estimate speed109.28 to 218.58 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153 GB/s divided by active weights in GB |
about 6.7 to 14 tokens/s
estimate, not measured by us
How we estimate speed6.67 to 13.96 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 90 to 180 tokens/s
estimate, not measured by us
How we estimate speed90 to 180 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153 GB/s divided by active weights in GB |
| MacBook Air 15 M5Apple M5 |
about 11 to 23 tokens/s
estimate, not measured by us
How we estimate speed10.97 to 22.95 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 109 to 219 tokens/s
estimate, not measured by us
How we estimate speed109.28 to 218.58 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153 GB/s divided by active weights in GB |
about 6.7 to 14 tokens/s
estimate, not measured by us
How we estimate speed6.67 to 13.96 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 90 to 180 tokens/s
estimate, not measured by us
How we estimate speed90 to 180 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153 GB/s divided by active weights in GB |
| MacBook Pro 14 M6Apple M6 |
about 11 to 26 tokens/s
estimate, not measured by us
How we estimate speed10.97 to 25.75 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153-170 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 109 to 243 tokens/s
estimate, not measured by us
How we estimate speed109.28 to 242.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
about 6.7 to 16 tokens/s
estimate, not measured by us
How we estimate speed6.67 to 15.66 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153-170 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 90 to 200 tokens/s
estimate, not measured by us
How we estimate speed90 to 200 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
| Mac mini M6Apple M6 |
about 11 to 26 tokens/s
estimate, not measured by us
How we estimate speed10.97 to 25.75 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153-170 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 109 to 243 tokens/s
estimate, not measured by us
How we estimate speed109.28 to 242.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
about 6.7 to 16 tokens/s
estimate, not measured by us
How we estimate speed6.67 to 15.66 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153-170 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 90 to 200 tokens/s
estimate, not measured by us
How we estimate speed90 to 200 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
| iMac M6Apple M6 |
about 11 to 26 tokens/s
estimate, not measured by us
How we estimate speed10.97 to 25.75 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153-170 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 71.7 tokens/s at 800 GB/s |
about 109 to 243 tokens/s
estimate, not measured by us
How we estimate speed109.28 to 242.86 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
about 6.7 to 16 tokens/s
estimate, not measured by us
How we estimate speed6.67 to 15.66 tokens/s; estimate: low end from memory bandwidth (Mac bandwidth 153-170 GB/s against 800 GB/s), high end from this chip's speed in llama.cpp's public Apple-silicon benchmark; measured on an M1 Ultra 64 GB: 43.6 tokens/s at 800 GB/s |
about 90 to 200 tokens/s
estimate, not measured by us
How we estimate speed90 to 200 tokens/s; estimate, not measured by us; 0.20 to 0.40 x Mac bandwidth 153-170 GB/s divided by active weights in GB |
What does not fit, and the next memory size
23 models exceed 16 GB for long work.
The next listed memory size is 18 GB. It admits 2 more models from this table. Those models are Mistral-7B-Instruct-v0.3, Meta-Llama-3.1-8B-Instruct. The last column below gives the next suitable size for each excluded model, including sizes outside these memory guides.
| Model | Short chats | Long 32,000-token work | Memory short for long work | Next listed memory |
|---|---|---|---|---|
| Mistral-7B-Instruct-v0.3 | 16 GB | 18 GB | 2 GB | 18 GB |
| Meta-Llama-3.1-8B-Instruct | 16 GB | 18 GB | 2 GB | 18 GB |
| Qwen3-8B | 16 GB | 24 GB | 8 GB | 24 GB |
| gemma-3-12b-it-qat | 24 GB | 48 GB | 32 GB | 48 GB |
| phi-4 | 24 GB | 32 GB | 16 GB | 32 GB |
| Qwen3-14B | 24 GB | 32 GB | 16 GB | 32 GB |
| gpt-oss-20b | 24 GB | 24 GB | 8 GB | 24 GB |
| Mistral-Small-3.2-24B-Instruct-2506 | 32 GB | 36 GB | 20 GB | 36 GB |
| Devstral-Small-2-24B-Instruct-2512 | 32 GB | 36 GB | 20 GB | 36 GB |
| Qwen3.6-27B | 24 GB measured | 24 GB measured | 8 GB | 24 GB |
| Qwen3.8-27B | 24 GB measured | 24 GB measured | 8 GB | 24 GB |
| gemma-3-27b-it-qat | 48 GB | 64 GB | 48 GB | 64 GB |
| Qwen3-30B-A3B | 32 GB | 36 GB | 20 GB | 36 GB |
| Qwen3-Coder-30B-A3B-Instruct | 32 GB | 36 GB | 20 GB | 36 GB |
| gemma-4-31b-it | 48 GB | 96 GB | 80 GB | 96 GB |
| DeepSeek-R1-Distill-Qwen-32B | 36 GB | 48 GB | 32 GB | 48 GB |
| Qwen3-32B | 36 GB | 48 GB | 32 GB | 48 GB |
| Qwen3.6-35B-A3B | 48 GB | 48 GB | 32 GB | 48 GB |
| Llama-3.3-70B-Instruct | 64 GB | 96 GB | 80 GB | 96 GB |
| Llama-4-Scout-17B-16E-Instruct | 96 GB | 128 GB | 112 GB | 128 GB |
| gpt-oss-120b | 96 GB | 96 GB | 80 GB | 96 GB |
| Qwen3-235B-A22B | 192 GB | 256 GB | 240 GB | 256 GB |
| Qwen3.5-397B-A17B | 64 GB measured | 64 GB measured | 48 GB | 64 GB |
Context length and data dates
Memory minimums use a 32K context; longer contexts need more memory. Default context is 32768 tokens.
3 fitting models have a M1 Ultra measurement. 7 models have rule-based minimums without our measurements. Figures as of 8 November 2026.
Mac source.
Apple technical specifications (apple.com/mac and the specs page of each Mac, read 2026-11-07; the iMac M6 page says only "up to 170GB/s", so its low end, 153 GB/s, is the 16 GB figure for the same M6 chip on the Mac mini page); table last changed 2026-11-07Chip ratio source.
llama.cpp's public Apple-silicon benchmark (github.com/ggml-org/llama.cpp/discussions/4167, read 2026-11-07): LLaMA 7B Q4_0 text generation, as a ratio to the M1 Ultra 48-core GPUModel source.
Dates come from each source text; otherwise table last changed 2026-10-19Reference test machine.
M1 Ultra, 64 GB, 800 GB/s| Model | Source date | Evidence |
|---|---|---|
| Qwen3.5-4B | 2026-04-29 | Measured on M1 Ultra, measured on 2026-04-29; the build details weren't recorded with this figure |
| Qwen3.5-9B | table last changed 2026-10-19 | Measured on M1 Ultra, the build and date weren't recorded with this figure |
| Gemma-4-26B-a4b-it | 2026-08-24 | Measured on M1 Ultra, from build 1.11.804 on 2026-08-24 |
| Qwen3.6-27B | table last changed 2026-10-19 | Measured on M1 Ultra, the build and date weren't recorded with this figure |
| Qwen3.5-397B-A17B | 2026-05-01 | Measured on M1 Ultra, measured on 2026-05-01; the build details weren't recorded with this figure |
| Qwen3.8-27B | 2026-08-19 | Measured on M1 Ultra, from build 1.11.792 on 2026-08-19 |
| Llama-3.3-70B-Instruct | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen2.5-Coder-7B-Instruct | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3-4B-Instruct-2507 | 2026-11-08 | Rule-based minimum, not measured by us |
| Devstral-Small-2-24B-Instruct-2512 | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3-0.6B | 2026-11-08 | Rule-based minimum, not measured by us |
| gemma-3-4b-it-qat | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3.6-35B-A3B | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3.5-0.8B | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen2.5-0.5B-Instruct | 2026-11-08 | Rule-based minimum, not measured by us |
| Llama-3.2-1B-Instruct | 2026-11-08 | Rule-based minimum, not measured by us |
| Meta-Llama-3.1-8B-Instruct | 2026-11-08 | Rule-based minimum, not measured by us |
| Mistral-7B-Instruct-v0.3 | 2026-11-08 | Rule-based minimum, not measured by us |
| Mistral-Small-3.2-24B-Instruct-2506 | 2026-11-08 | Rule-based minimum, not measured by us |
| gemma-3-12b-it-qat | 2026-11-08 | Rule-based minimum, not measured by us |
| gemma-3-27b-it-qat | 2026-11-08 | Rule-based minimum, not measured by us |
| gemma-4-31b-it | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3-8B | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3-14B | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3-32B | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3-Coder-30B-A3B-Instruct | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3-30B-A3B | 2026-11-08 | Rule-based minimum, not measured by us |
| gpt-oss-20b | 2026-11-08 | Rule-based minimum, not measured by us |
| gpt-oss-120b | 2026-11-08 | Rule-based minimum, not measured by us |
| DeepSeek-R1-Distill-Qwen-32B | 2026-11-08 | Rule-based minimum, not measured by us |
| phi-4 | 2026-11-08 | Rule-based minimum, not measured by us |
| Llama-4-Scout-17B-16E-Instruct | 2026-11-08 | Rule-based minimum, not measured by us |
| Qwen3-235B-A22B | 2026-11-08 | Rule-based minimum, not measured by us |
Outlier Pro: $39 once · Black Friday sale: $16.99 once through December 2. Then $39.
Free to start (Nano and Lite). macOS 15+, Apple silicon. Refund window: 30 days, no questions asked. Priced in US dollars. Checkout shows your currency and any local tax.
Try Outlier on your Mac
Use the memory table and the speed source labels above when choosing a local model.
Download Outlier