C58.5 tok/s decode
~$15,000 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
12.1 tok/s
TTFT
15972 ms
Safe context
23K
Memory
48.3 GB / 69.1 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 11.7 tok/s | 23970 ms | 40K |
| Chat | C | Runs well | 11.7 tok/s | 8989 ms | 12K |
| Coding | C | Runs well | 12.1 tok/s | 15972 ms | 23K |
| RAG | C | Runs well | 11.7 tok/s | 29963 ms | 40K |
| Reasoning | C | Runs well | 11.7 tok/s | 19476 ms | 23K |
How Qwen3 48B A4B Savant Commander Distill 12X Closed Open Heretic Uncensored (48B params) fits at each quantization level on MacBook Pro M4 Max 96GB (69.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 18.7 GB | Low | D35 |
Q3_K_S | 3 | 23.5 GB | Low | D36 |
NVFP4 |
Upgrade options
~$15,000 MSRP
~$30,000 MSRP
~$40,000 MSRP
| 4 |
26.9 GB |
| Medium |
| D37 |
Q4_K_M | 4 | 29.3 GB | Medium | D38 |
Q5_K_M | 5 | 34.6 GB | High | D40 |
Q6_K | 6 | 39.4 GB | High | C41 |
Q8_0Best for your GPU | 8 | 51.4 GB | Very High | C44 |
F16 | 16 | 98.4 GB | Maximum | F0 |