Can I run gemma-4-31B-it-Claude-Opus-Distill-v2 on a Apple M3 Pro?
Not at these settings. No indexed quantization of gemma-4-31B-it-Claude-Opus-Distill-v2 fits Apple M3 Pro at any context we compute, with q4_0 KV. The smallest shipped quantization is 12.82 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 | 57.20 GiB | 58.3 | 58.5 | 58.9 | 59.6 | 61.0 | 63.8 |
| Q8_0 | 30.39 GiB | 31.5 | 31.7 | 32.1 | 32.8 | 34.2 | 37.0 |
| Q6_K | 23.47 GiB | 24.6 | 24.8 | 25.1 | 25.8 | 27.2 | 30.1 |
| Q5_K_M | 20.35 GiB | 21.5 | 21.7 | 22.0 | 22.7 | 24.1 | 26.9 |
| Q4_K_M | 17.40 GiB | 18.5 | 18.7 | 19.1 | 19.8 | 21.2 | 24.0 |
| IQ4_NL | 16.53 GiB | 17.7 | 17.8 | 18.2 | 18.9 | 20.3 | 23.1 |
| Q3_K_M | 14.24 GiB | 15.4 | 15.5 | 15.9 | 16.6 | 18.0 | 20.8 |
| Q3_K_S | 12.82 GiB | 14.0 | 14.1 | 14.5 | 15.2 | 16.6 | 19.4 |
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 1,024-token window rather than the full context.