Can I run Llama-4-Maverick-17B-128E-Instruct on a Apple M5 Max?
Not at these settings. No indexed quantization of Llama-4-Maverick-17B-128E-Instruct fits Apple M5 Max at any context we compute, with q4_0 KV. The smallest shipped quantization is 92.59 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 | 746.43 GiB | 747.2 | 747.4 | 747.8 | 748.7 | 750.4 | 753.8 |
| Q8_0 | 396.57 GiB | 397.4 | 397.6 | 398.0 | 398.8 | 400.5 | 403.9 |
| Q6_K | 306.19 GiB | 307.0 | 307.2 | 307.6 | 308.5 | 310.1 | 313.5 |
| UD-IQ2_M | 273.65 GiB | 274.4 | 274.6 | 275.1 | 275.9 | 277.6 | 281.0 |
| Q5_K_M | 264.93 GiB | 265.7 | 265.9 | 266.4 | 267.2 | 268.9 | 272.3 |
| Q5_K_S | 256.76 GiB | 257.5 | 257.8 | 258.2 | 259.0 | 260.7 | 264.1 |
| Q4_1 | 233.50 GiB | 234.3 | 234.5 | 234.9 | 235.8 | 237.4 | 240.8 |
| Q4_K_M | 226.09 GiB | 226.9 | 227.1 | 227.5 | 228.4 | 230.0 | 233.4 |
| Q4_K_S | 212.15 GiB | 212.9 | 213.2 | 213.6 | 214.4 | 216.1 | 219.5 |
| Q4_0 | 211.19 GiB | 212.0 | 212.2 | 212.6 | 213.5 | 215.1 | 218.5 |
| IQ4_NL | 210.26 GiB | 211.0 | 211.3 | 211.7 | 212.5 | 214.2 | 217.6 |
| UD-IQ4_XS | 205.52 GiB | 206.3 | 206.5 | 206.9 | 207.8 | 209.5 | 212.8 |
| IQ4_XS | 199.61 GiB | 200.4 | 200.6 | 201.0 | 201.9 | 203.6 | 206.9 |
| Q3_K_M | 177.95 GiB | 178.7 | 178.9 | 179.4 | 180.2 | 181.9 | 185.3 |
| Q3_K_S | 160.80 GiB | 161.6 | 161.8 | 162.2 | 163.1 | 164.7 | 168.1 |
| UD-IQ3_XXS | 157.69 GiB | 158.5 | 158.7 | 159.1 | 160.0 | 161.6 | 165.0 |
| Q2_K_L | 135.87 GiB | 136.7 | 136.9 | 137.3 | 138.1 | 139.8 | 143.2 |
| Q2_K | 135.64 GiB | 136.4 | 136.6 | 137.1 | 137.9 | 139.6 | 143.0 |
| UD-IQ2_XXS | 125.91 GiB | 126.7 | 126.9 | 127.3 | 128.2 | 129.9 | 133.2 |
| UD-IQ1_M | 118.78 GiB | 119.6 | 119.8 | 120.2 | 121.0 | 122.7 | 126.1 |
| UD-IQ1_S | 112.48 GiB | 113.3 | 113.5 | 113.9 | 114.7 | 116.4 | 119.8 |
| UD-TQ1_0 | 98.45 GiB | 99.2 | 99.5 | 99.9 | 100.7 | 102.4 | 105.8 |
| TQ1_0 | 92.59 GiB | 93.4 | 93.6 | 94.0 | 94.9 | 96.5 | 99.9 |
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 8,192-token window rather than the full context.