Can I run gpt-oss-120b-uncensored-bf16 on a Apple M5 Max?
Not at these settings. No indexed quantization of gpt-oss-120b-uncensored-bf16 fits Apple M5 Max at any context we compute, with q8_0 KV. The smallest shipped quantization is 58.30 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◐ |
|---|---|---|---|---|---|---|---|
| Q6_K | 58.94 GiB | 59.6 | 59.6 | 59.8 | 60.1 | 60.7 | 61.9 |
| Q4_K_L | 58.73 GiB | 59.4 | 59.4 | 59.6 | 59.9 | 60.5 | 61.7 |
| Q2_K_L | 58.67 GiB | 59.3 | 59.4 | 59.5 | 59.8 | 60.4 | 61.6 |
| Q4_K_M | 58.53 GiB | 59.2 | 59.2 | 59.4 | 59.7 | 60.3 | 61.5 |
| IQ4_NL | 58.40 GiB | 59.0 | 59.1 | 59.3 | 59.6 | 60.2 | 61.3 |
| IQ3_M | 58.40 GiB | 59.0 | 59.1 | 59.3 | 59.6 | 60.1 | 61.3 |
| Q2_K | 58.40 GiB | 59.0 | 59.1 | 59.3 | 59.6 | 60.1 | 61.3 |
| IQ2_M | 58.38 GiB | 59.0 | 59.1 | 59.2 | 59.5 | 60.1 | 61.3 |
| Q3_K_L | 58.30 GiB | 58.9 | 59.0 | 59.2 | 59.5 | 60.0 | 61.2 |
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.