Can I run MiMo-V2.5 on a Apple M5 Max?
Not at these settings. No indexed quantization of MiMo-V2.5 fits Apple M5 Max at any context we compute, with q4_0 KV. The smallest shipped quantization is 60.09 GiB in weights alone, against 44.64 GiB usable. CPU offload can still run it, slowly.
Every quantization at every context
| Quant | Weights● | 4K◐ | 8K◐ | 16K◐ | 32K◐ | 64K◐ | 128K◐ |
|---|---|---|---|---|---|---|---|
| BF16 | 577.08 GiB | 577.8 | 577.9 | 578.2 | 578.7 | 579.8 | 581.9 |
| Q8_0 | 306.67 GiB | 307.4 | 307.5 | 307.8 | 308.3 | 309.4 | 311.5 |
| Q6_K | 249.18 GiB | 249.9 | 250.0 | 250.3 | 250.8 | 251.9 | 254.0 |
| UD-Q6_K | 239.34 GiB | 240.1 | 240.2 | 240.5 | 241.0 | 242.0 | 244.2 |
| UD-Q5_K_M | 214.37 GiB | 215.1 | 215.2 | 215.5 | 216.0 | 217.1 | 219.2 |
| Q5_K_M | 206.24 GiB | 207.0 | 207.1 | 207.4 | 207.9 | 208.9 | 211.1 |
| UD-Q5_K_S | 201.45 GiB | 202.2 | 202.3 | 202.6 | 203.1 | 204.2 | 206.3 |
| Q5_K_S | 199.54 GiB | 200.3 | 200.4 | 200.7 | 201.2 | 202.2 | 204.4 |
| Q4_1 | 180.99 GiB | 181.7 | 181.8 | 182.1 | 182.6 | 183.7 | 185.8 |
| UD-Q4_K_M | 177.81 GiB | 178.5 | 178.7 | 178.9 | 179.5 | 180.5 | 182.6 |
| Q4_K_L | 176.24 GiB | 177.0 | 177.1 | 177.4 | 177.9 | 178.9 | 181.1 |
| Q4_K_M | 175.80 GiB | 176.5 | 176.7 | 176.9 | 177.5 | 178.5 | 180.6 |
| Q4_K_S | 169.41 GiB | 170.1 | 170.3 | 170.5 | 171.1 | 172.1 | 174.2 |
| UD-Q4_K_S | 166.56 GiB | 167.3 | 167.4 | 167.7 | 168.2 | 169.3 | 171.4 |
| Q4_K | 165.75 GiB | 166.5 | 166.6 | 166.9 | 167.4 | 168.5 | 170.6 |
| Q4_0 | 163.68 GiB | 164.4 | 164.5 | 164.8 | 165.3 | 166.4 | 168.5 |
| IQ4_NL | 163.24 GiB | 164.0 | 164.1 | 164.4 | 164.9 | 165.9 | 168.1 |
| IQ4_XS | 154.32 GiB | 155.0 | 155.2 | 155.4 | 156.0 | 157.0 | 159.1 |
| UD-IQ4_NL | 142.12 GiB | 142.8 | 143.0 | 143.2 | 143.8 | 144.8 | 146.9 |
| UD-IQ4_XS | 139.18 GiB | 139.9 | 140.0 | 140.3 | 140.8 | 141.9 | 144.0 |
| IQ3_M | 138.15 GiB | 138.9 | 139.0 | 139.3 | 139.8 | 140.9 | 143.0 |
| Q3_K_L | 137.77 GiB | 138.5 | 138.6 | 138.9 | 139.4 | 140.5 | 142.6 |
| Q3_K_M | 132.47 GiB | 133.2 | 133.3 | 133.6 | 134.1 | 135.2 | 137.3 |
| IQ3_XS | 132.46 GiB | 133.2 | 133.3 | 133.6 | 134.1 | 135.2 | 137.3 |
| UD-Q3_K_M | 130.41 GiB | 131.1 | 131.3 | 131.5 | 132.1 | 133.1 | 135.2 |
| Q3_K_S | 125.83 GiB | 126.6 | 126.7 | 127.0 | 127.5 | 128.5 | 130.6 |
| IQ3_XXS | 121.31 GiB | 122.0 | 122.2 | 122.4 | 123.0 | 124.0 | 126.1 |
| UD-IQ3_XXS | 117.27 GiB | 118.0 | 118.1 | 118.4 | 118.9 | 120.0 | 122.1 |
| UD-IQ3_S | 106.98 GiB | 107.7 | 107.8 | 108.1 | 108.6 | 109.7 | 111.8 |
| IQ3_S | 106.20 GiB | 106.9 | 107.1 | 107.3 | 107.9 | 108.9 | 111.0 |
| Q2_K_L | 102.02 GiB | 102.8 | 102.9 | 103.2 | 103.7 | 104.7 | 106.8 |
| Q2_K | 101.46 GiB | 102.2 | 102.3 | 102.6 | 103.1 | 104.2 | 106.3 |
| IQ2_M | 97.30 GiB | 98.0 | 98.2 | 98.4 | 99.0 | 100.0 | 102.1 |
| UD-IQ2_M | 89.93 GiB | 90.7 | 90.8 | 91.1 | 91.6 | 92.6 | 94.7 |
| UD-IQ2_XXS | 89.85 GiB | 90.6 | 90.7 | 91.0 | 91.5 | 92.6 | 94.7 |
| IQ2_S | 88.18 GiB | 88.9 | 89.0 | 89.3 | 89.8 | 90.9 | 93.0 |
| IQ2_XS | 86.68 GiB | 87.4 | 87.5 | 87.8 | 88.3 | 89.4 | 91.5 |
| UD-IQ1_M | 86.17 GiB | 86.9 | 87.0 | 87.3 | 87.8 | 88.9 | 91.0 |
| IQ2_XXS | 79.83 GiB | 80.6 | 80.7 | 81.0 | 81.5 | 82.5 | 84.6 |
| IQ1_M | 67.01 GiB | 67.7 | 67.9 | 68.1 | 68.7 | 69.7 | 71.8 |
| IQ1_S | 60.09 GiB | 60.8 | 60.9 | 61.2 | 61.7 | 62.8 | 64.9 |
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 128-token window rather than the full context.