M4 Max 128GB vs M5 32GB for local LLMs
12 reports on the M4 Max 128GB and 3 on the M5 32GB
Which is faster for local LLMs?
On Qwen3 30B · 3B active at 4-bit, the M4 Max 128GB runs at 127.7 tokens per second (one run) and the M5 32GB at 57.0 (one run). Those are community reports whose engines and settings differ, so the gap is a rough sign, not a controlled measure of the two cards. Memory bandwidth, which caps how fast a card can write, is 3.6x higher on the M4 Max 128GB (546 GB/s on the M4 Max 128GB, 154 on the M5 32GB).
- Memory
- 128GB unified
- Memory bandwidth
- 546 GB/s
- FP16 compute
- 18.4 TFLOPS
- Reports
- 12
- Memory
- 32GB unified
- Memory bandwidth
- 154 GB/s
- FP16 compute
- not on record
- Reports
- 3
Measured on both
Median tokens per second from plain runs (one device, one request, no speculative decoding, the whole model in the device's memory) of the same model size at the same quant level. Engines and context lengths can still differ between the runs. How to read these
| Model | M4 Max 128GB | M5 32GB |
|---|---|---|
| Qwen3 30B · 3B active4-bit | 127.7one run | 57.0one run |
Estimated by the calculator
The quant the calculator recommends for each card and its estimated speed at a 32,768-token context with 32 GB of system RAM, from memory bandwidth and the model's shape. Where a model was also measured above, the measurement wins.
| Model | M4 Max 128GB | M5 32GB |
|---|---|---|
| Qwen3 30B · 3B active | 23.7 t/sFP16 | 53.0 t/sQ2_K |
| Qwen3.8 27B | 6.7 t/sFP16 | 9.2 t/sQ4_K_S |
| Qwen3.8 125B · 6B active | 30.7 t/sQ5_K_S | doesn't fit at these settings |
| DeepSeek V4 Flash 284B · 13B active | 48.3 t/sQ2_K | doesn't fit at these settings |
| Qwen3.6 35B · 3B active | 21.2 t/sFP16 | 33.5 t/sQ3_K_M |
| Qwen3.6 27B | 6.7 t/sFP16 | 9.2 t/sQ4_K_S |
| Gemma 4 26B · 4B active | 18.2 t/sFP16 | 20.3 t/sQ4_K_M |
Change the context, RAM or model in the calculator: M4 Max 128GB · M5 32GB
Every report on each card, with its source: M4 Max 128GB · M5 32GB