What can the Apple M5 Max (40-core GPU, 128GB) run?
128 GB unified memory (~96.0 GB usable for models) · 614 GB/s memory bandwidth
Measured: 119.92 tok/s on Llama 2 7B Q4_0 (llama.cpp tg128). Source: llama.cpp Apple Silicon benchmarks (Llama 2 7B Q4_0, tg128)
| Model | Verdict | Best quant that fits | Est. tok/s (Q4) |
|---|---|---|---|
| Llama 3.1 8B Instruct | Yes | F16 | ~94.5 |
| Llama 3.3 70B Instruct | Yes, usable | Q8_0 | ~10.7 |
| Llama 3.2 3B Instruct | Yes | F16 | ~236 |
| Llama 4 Scout (17B active, 16 experts) | Yes | Q6_K | ~26.9 |
| Qwen3 8B | Yes | F16 | ~92.2 |
| Qwen3 14B | Yes | F16 | ~51.1 |
| Qwen3 32B | Yes | F16 | ~23.1 |
| Qwen3 30B-A3B (MoE) | Yes | F16 | ~113 |
| Qwen3 235B-A22B (MoE) | No | — | — |
| Qwen2.5 Coder 32B Instruct | Yes | F16 | ~23.1 |
| Gemma 3 4B | Yes | F16 | ~176 |
| Gemma 3 12B | Yes | F16 | ~62.0 |
| Gemma 3 27B | Yes | F16 | ~27.6 |
| Mistral Small 3.1 24B | Yes | F16 | ~31.5 |
| Mistral Nemo 12B | Yes | F16 | ~62.0 |
| Phi-4 14B | Yes | F16 | ~51.4 |
| DeepSeek R1 Distill Qwen 32B | Yes | F16 | ~23.1 |
| DeepSeek R1 Distill Llama 70B | Yes, usable | Q8_0 | ~10.7 |
| gpt-oss-20b (MoE) | Yes | MXFP4 (native) | ~110 |
| gpt-oss-120b (MoE) | Yes | MXFP4 (native) | ~65.1 |
| DeepSeek V3 / R1 671B (MoE) | No | — | — |
Estimated tok/s = calibrated bandwidth ÷ active weight size. Real results vary with runtime, context and drivers.
No suitable GPU? Rent one by the hour on Vast.ai or RunPod (affiliate links).