Can I run Kimi-Dev-72B on a Apple M3 Pro?
Not at these settings. No indexed quantization of Kimi-Dev-72B fits Apple M3 Pro at any context we compute, with q4_0 KV. The smallest shipped quantization is 21.47 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◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 135.44 GiB | 136.5 | 136.8 | 137.5 | 138.9 | 141.7 | 147.4 |
| Q8_0 | 71.96 GiB | 73.0 | 73.3 | 74.0 | 75.4 | 78.3 | 83.9 |
| Q6_K | 59.93 GiB | 61.0 | 61.3 | 62.0 | 63.4 | 66.2 | 71.9 |
| Q4_1 | 42.56 GiB | 43.6 | 43.9 | 44.6 | 46.1 | 48.9 | 54.5 |
| Q4_0 | 38.54 GiB | 39.6 | 39.9 | 40.6 | 42.0 | 44.8 | 50.5 |
| IQ4_NL | 38.48 GiB | 39.5 | 39.9 | 40.6 | 42.0 | 44.8 | 50.4 |
| IQ4_XS | 37.02 GiB | 38.0 | 38.4 | 39.1 | 40.5 | 43.3 | 48.9 |
| Q3_K_M | 35.11 GiB | 36.1 | 36.5 | 37.2 | 38.6 | 41.4 | 47.0 |
| Q3_K_S | 32.12 GiB | 33.2 | 33.5 | 34.2 | 35.6 | 38.4 | 44.0 |
| UD-IQ3_XXS | 29.67 GiB | 30.7 | 31.1 | 31.8 | 33.2 | 36.0 | 41.6 |
| UD-IQ2_M | 27.56 GiB | 28.6 | 28.9 | 29.7 | 31.1 | 33.9 | 39.5 |
| UD-IQ2_XXS | 23.94 GiB | 25.0 | 25.3 | 26.0 | 27.4 | 30.2 | 35.9 |
| UD-IQ1_M | 22.35 GiB | 23.4 | 23.7 | 24.4 | 25.8 | 28.7 | 34.3 |
| UD-IQ1_S | 21.47 GiB | 22.5 | 22.8 | 23.6 | 25.0 | 27.8 | 33.4 |
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 131,072-token window rather than the full context.