Can I run command-a-plus-05-2026-bf16 on a Apple M3 Pro?
Not at these settings. No indexed quantization of command-a-plus-05-2026-bf16 fits Apple M3 Pro at any context we compute, with q4_0 KV. The smallest shipped quantization is 45.87 GiB in weights alone, against 12.56 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| Q8_0 | 216.03 GiB | 216.7 | 216.8 | 216.8 | 217.0 | 217.3 | 217.8 |
| Q6_K | 176.79 GiB | 177.5 | 177.5 | 177.6 | 177.7 | 178.0 | 178.6 |
| Q5_K_M | 146.93 GiB | 147.6 | 147.7 | 147.7 | 147.9 | 148.2 | 148.7 |
| Q5_K_S | 141.69 GiB | 142.4 | 142.4 | 142.5 | 142.6 | 142.9 | 143.5 |
| Q4_1 | 129.29 GiB | 130.0 | 130.0 | 130.1 | 130.2 | 130.5 | 131.1 |
| Q4_K_L | 126.06 GiB | 126.8 | 126.8 | 126.9 | 127.0 | 127.3 | 127.9 |
| Q4_K_M | 125.81 GiB | 126.5 | 126.6 | 126.6 | 126.8 | 127.1 | 127.6 |
| Q4_K_S | 120.92 GiB | 121.6 | 121.7 | 121.7 | 121.9 | 122.2 | 122.7 |
| Q4_0 | 117.42 GiB | 118.1 | 118.2 | 118.2 | 118.4 | 118.7 | 119.2 |
| IQ4_NL | 116.79 GiB | 117.5 | 117.5 | 117.6 | 117.7 | 118.0 | 118.6 |
| IQ4_XS | 110.67 GiB | 111.4 | 111.4 | 111.5 | 111.6 | 111.9 | 112.5 |
| IQ3_M | 99.41 GiB | 100.1 | 100.2 | 100.2 | 100.4 | 100.6 | 101.2 |
| Q3_K_L | 99.27 GiB | 100.0 | 100.0 | 100.1 | 100.2 | 100.5 | 101.1 |
| Q3_K_M | 95.53 GiB | 96.2 | 96.3 | 96.3 | 96.5 | 96.8 | 97.3 |
| IQ3_XS | 95.16 GiB | 95.9 | 95.9 | 96.0 | 96.1 | 96.4 | 97.0 |
| Q3_K_S | 91.28 GiB | 92.0 | 92.0 | 92.1 | 92.2 | 92.5 | 93.1 |
| IQ3_XXS | 87.48 GiB | 88.2 | 88.2 | 88.3 | 88.4 | 88.7 | 89.3 |
| Q2_K_L | 74.80 GiB | 75.5 | 75.5 | 75.6 | 75.8 | 76.0 | 76.6 |
| Q2_K | 74.56 GiB | 75.3 | 75.3 | 75.4 | 75.5 | 75.8 | 76.4 |
| IQ2_M | 71.32 GiB | 72.0 | 72.1 | 72.1 | 72.3 | 72.6 | 73.1 |
| IQ2_S | 65.10 GiB | 65.8 | 65.8 | 65.9 | 66.1 | 66.3 | 66.9 |
| IQ2_XS | 63.98 GiB | 64.7 | 64.7 | 64.8 | 64.9 | 65.2 | 65.8 |
| IQ2_XXS | 57.98 GiB | 58.7 | 58.7 | 58.8 | 58.9 | 59.2 | 59.8 |
| IQ1_M | 50.59 GiB | 51.3 | 51.3 | 51.4 | 51.5 | 51.8 | 52.4 |
| IQ1_S | 45.87 GiB | 46.6 | 46.6 | 46.7 | 46.8 | 47.1 | 47.7 |
Figures are GiB of total memory: weights plus KV cache plus compute buffer and backend overhead. Weights and KV are near-exact; the overhead term is modeled. Hover any cell for the breakdown.
Why other calculators disagree
A parameters × bits ÷ 8 estimate ignores two things that dominate at long context. First, the weights themselves are not the nominal rate — quantizations are mixtures, so the real file is consistently larger than the label implies. Second, most of this model's layers cache only a 4,096-token window rather than the full context.