Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
213.1 tok/s
TTFT
908 ms
Safe context
65K
Memory
11.4 GB / 46.1 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 190.2 tok/s | 1481 ms | 125K |
| Chat | C | Runs well | 190.2 tok/s | 555 ms | 32K |
| Coding | C | Runs well | 213.1 tok/s | 908 ms | 65K |
| RAG | C | Runs well | 190.2 tok/s | 1851 ms | 125K |
| Reasoning | C | Runs well | 190.2 tok/s | 1203 ms | 65K |
How Qwen3.5 4B (4B params) fits at each quantization level on Mac Studio M2 Ultra 64GB (46.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 1.6 GB | Low | D30 |
Q3_K_S | 3 | 2.0 GB | Low | D30 |
NVFP4 | 4 |
2.2 GB |
| Medium |
| D31 |
Q4_K_M | 4 | 2.4 GB | Medium | D31 |
Q5_K_M | 5 | 2.9 GB | High | D31 |
Q6_K | 6 | 3.3 GB | High | D31 |
Q8_0 | 8 | 4.3 GB | Very High | D31 |
F16Best for your GPU | 16 | 8.2 GB | Maximum | D33 |