Can I run MiMo-V2.5-Pro on a Apple M5 Max?
Not at these settings. No indexed quantization of MiMo-V2.5-Pro fits Apple M5 Max at any context we compute, with q8_0 KV. The smallest shipped quantization is 197.49 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 | 1906.14 GiB | 1907.5 | 1908.2 | 1909.7 | 1912.6 | 1918.4 | 1930.0 |
| Q8_0 | 1012.92 GiB | 1014.3 | 1015.0 | 1016.4 | 1019.3 | 1025.2 | 1036.8 |
| UD-Q6_K | 788.42 GiB | 789.8 | 790.5 | 791.9 | 794.8 | 800.7 | 812.3 |
| UD-Q5_K_M | 705.94 GiB | 707.3 | 708.0 | 709.5 | 712.4 | 718.2 | 729.8 |
| Q5_K_M | 704.85 GiB | 706.2 | 706.9 | 708.4 | 711.3 | 717.1 | 728.7 |
| UD-Q5_K_S | 664.21 GiB | 665.5 | 666.3 | 667.7 | 670.6 | 676.4 | 688.1 |
| Q4_1 | 596.26 GiB | 597.6 | 598.3 | 599.8 | 602.7 | 608.5 | 620.1 |
| UD-Q4_K_M | 586.37 GiB | 587.7 | 588.4 | 589.9 | 592.8 | 598.6 | 610.2 |
| Q4_K_M | 585.99 GiB | 587.3 | 588.1 | 589.5 | 592.4 | 598.2 | 609.8 |
| Q4_K_S | 557.32 GiB | 558.7 | 559.4 | 560.8 | 563.7 | 569.6 | 581.2 |
| UD-Q4_K_S | 548.12 GiB | 549.5 | 550.2 | 551.6 | 554.5 | 560.4 | 572.0 |
| Q4_0 | 539.02 GiB | 540.4 | 541.1 | 542.5 | 545.4 | 551.3 | 562.9 |
| IQ4_NL | 537.62 GiB | 539.0 | 539.7 | 541.1 | 544.0 | 549.9 | 561.5 |
| IQ4_XS | 508.09 GiB | 509.4 | 510.2 | 511.6 | 514.5 | 520.3 | 531.9 |
| UD-IQ4_NL | 466.84 GiB | 468.2 | 468.9 | 470.4 | 473.3 | 479.1 | 490.7 |
| UD-IQ4_XS | 457.00 GiB | 458.3 | 459.1 | 460.5 | 463.4 | 469.2 | 480.9 |
| IQ3_M | 454.54 GiB | 455.9 | 456.6 | 458.1 | 461.0 | 466.8 | 478.4 |
| Q3_K_L | 452.86 GiB | 454.2 | 454.9 | 456.4 | 459.3 | 465.1 | 476.7 |
| Q3_K_M | 434.60 GiB | 435.9 | 436.7 | 438.1 | 441.0 | 446.8 | 458.5 |
| IQ3_XS | 434.58 GiB | 435.9 | 436.6 | 438.1 | 441.0 | 446.8 | 458.4 |
| UD-Q3_K_M | 428.10 GiB | 429.4 | 430.2 | 431.6 | 434.5 | 440.3 | 452.0 |
| Q3_K_S | 414.26 GiB | 415.6 | 416.3 | 417.8 | 420.7 | 426.5 | 438.1 |
| IQ3_XXS | 397.96 GiB | 399.3 | 400.0 | 401.5 | 404.4 | 410.2 | 421.8 |
| UD-IQ3_XXS | 384.32 GiB | 385.7 | 386.4 | 387.8 | 390.7 | 396.6 | 408.2 |
| UD-IQ3_S | 351.94 GiB | 353.3 | 354.0 | 355.5 | 358.4 | 364.2 | 375.8 |
| IQ3_S | 350.82 GiB | 352.2 | 352.9 | 354.3 | 357.2 | 363.1 | 374.7 |
| Q2_K_L | 334.58 GiB | 335.9 | 336.6 | 338.1 | 341.0 | 346.8 | 358.4 |
| Q2_K | 333.73 GiB | 335.1 | 335.8 | 337.2 | 340.2 | 346.0 | 357.6 |
| IQ2_M | 321.02 GiB | 322.4 | 323.1 | 324.5 | 327.4 | 333.3 | 344.9 |
| IQ2_S | 297.46 GiB | 298.8 | 299.5 | 301.0 | 303.9 | 309.7 | 321.3 |
| UD-IQ2_M | 295.48 GiB | 296.8 | 297.5 | 299.0 | 301.9 | 307.7 | 319.3 |
| UD-IQ2_XXS | 295.37 GiB | 296.7 | 297.4 | 298.9 | 301.8 | 307.6 | 319.2 |
| IQ2_XS | 285.38 GiB | 286.7 | 287.4 | 288.9 | 291.8 | 297.6 | 309.2 |
| UD-IQ1_M | 283.21 GiB | 284.5 | 285.3 | 286.7 | 289.6 | 295.4 | 307.1 |
| IQ2_XXS | 256.27 GiB | 257.6 | 258.3 | 259.8 | 262.7 | 268.5 | 280.1 |
| IQ1_M | 220.44 GiB | 221.8 | 222.5 | 224.0 | 226.9 | 232.7 | 244.3 |
| IQ1_S | 197.49 GiB | 198.8 | 199.6 | 201.0 | 203.9 | 209.7 | 221.3 |
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.