# M4 Max (14-core CPU / 32-core GPU)

> OnPrem World · https://onpremworld.com/en/hardware/apple-m4-max-32c-gpu · Last updated: 2026-10-14 · CC BY 4.0

The binned M4 Max: 410 GB/s but fixed at 36 GB, enough for 32B models at 5-bit.

## Specifications

|  |  |
|---|---|
| Memory | 36 GB LPDDR5X |
| Memory bandwidth | 410 GB/s |
| Launch | 2024-11 |
| Launch price | US$3,199 |
| Architecture | Apple M4 family (second-generation 3 nm; TSMC N3E) |
| Software | Metal / MLX |

## What it can run (36 GB, 8K)

Estimates, not benchmarks. See the methodology.

| Model | Parameters | Best quant | Total | tok/s (est.) |
|---|---|---|---|---|
| Qwen3.5-35B-A3B | 35B | Q5_K_M | 25.0 GiB | 125 |
| Qwen3.6-35B-A3B | 35B | Q5_K_M | 25.0 GiB | 125 |
| Qwen3-32B | 32.8B | Q5_K_M | 25.3 GiB | 11 |
| DeepSeek-R1-Distill-Qwen-32B | 32.76B | Q5_K_M | 25.3 GiB | 11 |
| Gemma 4 31B IT | 30.7B | Q6_K | 26.3 GiB | 11 |
| Qwen3-30B-A3B (2507) | 30.5B | Q6_K | 25.8 GiB | 98 |
| Qwen3-Coder-30B-A3B-Instruct | 30.5B | Q6_K | 25.8 GiB | 98 |
| GLM-4.7-Flash (30B-A3B) | 30B | Q6_K | 25.1 GiB | 108 |
| Qwen3.5-27B | 27B | Q6_K | 22.7 GiB | 12 |
| Qwen3.6-27B | 27B | Q6_K | 22.7 GiB | 12 |
| Qwen3.8-27B | 27B | Q6_K | 22.7 GiB | 12 |
| Gemma 3 27B IT | 27B | Q6_K | 23.3 GiB | 12 |
| Gemma 4 26B A4B IT | 25.2B | Q8_0 | 27.0 GiB | 66 |
| Mistral Small 3.2 24B Instruct (2506) | 24B | Q8_0 | 26.7 GiB | 10 |
| Magistral Small 1.2 (2509) | 24B | Q8_0 | 26.7 GiB | 10 |
| Devstral Small 2 24B Instruct (2512) | 24B | Q8_0 | 26.7 GiB | 10 |
| gpt-oss-20b | 21B | Q8_0 | 22.5 GiB | 70 |
| Phi-4 (14B) | 14B | Q8_0 | 16.6 GiB | 18 |
| Gemma 3 12B IT | 12B | FP16 | 24.8 GiB | 11 |
| Gemma 4 12B Unified IT | 11.95B | FP16 | 24.2 GiB | 11 |
| Qwen3.5-9B | 9B | FP16 | 18.4 GiB | 15 |
| Qwen3-8B | 8.2B | FP16 | 17.7 GiB | 16 |
| DeepSeek-R1-0528-Qwen3-8B | 8.19B | FP16 | 17.6 GiB | 16 |
| Llama 3.1 8B Instruct | 8B | FP16 | 17.1 GiB | 17 |
| Gemma 4 E4B IT | 8B | FP16 | 16.4 GiB | 17 |
| Llama 3.2 3B Instruct | 3.21B | FP16 | 7.7 GiB | 42 |
| Kimi K2.6 | 1000B | Doesn't fit | – | – |
| DeepSeek-R1-0528 (671B) | 671B | Doesn't fit | – | – |
| DeepSeek-V4-Flash | 284B | Doesn't fit | – | – |
| Qwen3.8-Flash-Next | 180B | Doesn't fit | – | – |
| Qwen3.5-122B-A10B | 122B | Doesn't fit | – | – |
| Mistral Small 4 119B (2603) | 119B | Doesn't fit | – | – |
| gpt-oss-120b | 117B | Doesn't fit | – | – |
| Llama 4 Scout 17B-16E Instruct | 109B | Doesn't fit | – | – |
| GLM-4.5-Air | 106B | Doesn't fit | – | – |
| Qwen3-Coder-Next (80B-A3B) | 80B | Doesn't fit | – | – |
| Llama 3.3 70B Instruct | 70B | Doesn't fit | – | – |

## Prices

No prices recorded for this device yet.

## Frequently asked questions

### Which LLMs can the M4 Max (14-core CPU / 32-core GPU) run?

With 36 GB and an 8K context, the largest models that fit include: Qwen3.5-35B-A3B, Qwen3.6-35B-A3B, Qwen3-32B, DeepSeek-R1-Distill-Qwen-32B, Gemma 4 31B IT. See the full table below.

### How fast is the M4 Max (14-core CPU / 32-core GPU) for local LLMs?

Generation speed is mostly set by memory bandwidth (410 GB/s). For Llama 3.1 8B Instruct at FP16, expect roughly 17 tokens/s; bigger models are slower.

### Can the M4 Max (14-core CPU / 32-core GPU) run 70B models?

Llama 3.3 70B Instruct does not fit on the M4 Max (14-core CPU / 32-core GPU) (36 GB), even at Q4_K_M with an 8K context.

## Sources

- https://support.apple.com/en-us/121553
- https://en.wikipedia.org/wiki/Apple_M4
- https://www.apple.com/newsroom/2024/10/new-macbook-pro-features-m4-family-of-chips-and-apple-intelligence/
- https://web.archive.org/web/20241121061833/https://www.apple.com/shop/buy-mac/macbook-pro/14-inch-m4-max
- https://www.apple.com/newsroom/2024/10/apple-introduces-m4-pro-and-m4-max/