Apple · M5 Pro · Mini PC
Best AI Models for Mac Mini M5 Pro (24 GB)
Runs 14B models well and fits 35B MoE; dense 27–31B fits but is slow at 12 tok/s, and 70B doesn't fit in 20.5 GB.
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB | 12.3 tok/s | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB | 12.3 tok/s | |
| 3 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB | 53.2 tok/s | |
| 4 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB | 59.3 tok/s | |
| 5 | MiniCPM5 2BOpenBMB Q4_K_M · 2.5 GB | 113 tok/s |
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Q4_K_M · 16.5 GB | 53.2 tok/s | |
| 2 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB | 59.3 tok/s | |
| 3 | Q4_K_M · 7.9 GB | 40.5 tok/s | |
| 4 | Qwen3.8 27BAlibaba Q4_K_M · 17.9 GB | 12.3 tok/s | |
| 5 | MiniCPM5 2BOpenBMB Q4_K_M · 2.5 GB | 113 tok/s |
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Q4_K_M · 16.5 GB | 53.2 tok/s | |
| 2 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB | 33.8 tok/s | |
| 3 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB | 59.3 tok/s | |
| 4 | Qwen3 8BAlibaba Q4_K_M · 7.6 GB | 38.9 tok/s | |
| 5 | MiniCPM5 2BOpenBMB Q4_K_M · 2.5 GB | 113 tok/s |
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB | 33.8 tok/s | |
| 2 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB | 59.3 tok/s | |
| 3 | Mistral Nemo Instruct 2407Mistral AI Q4_K_M · 11.0 GB | 26.6 tok/s | |
| 4 | Q4_K_M · 7.9 GB | 40.5 tok/s | |
| 5 | MiniCPM5 2BOpenBMB Q4_K_M · 2.5 GB | 113 tok/s |
| # | Model | Speed ↓ | Score |
|---|---|---|---|
| 1 | GPT OSS 20BOpenAIHow to run → Q4_K_M · 13.1 GB | 59.3 tok/s | |
| 2 | MiniCPM5 2BOpenBMB Q4_K_M · 2.5 GB | 113 tok/s | |
| 3 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB | 53.2 tok/s | |
| 4 | Qwen3 8BAlibaba Q4_K_M · 7.6 GB | 38.9 tok/s | |
| 5 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB | 33.8 tok/s |
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Q4_K_M · 16.5 GB | 53.2 tok/s | |
| 2 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB | 33.8 tok/s | |
| 3 | Gemma 3 12B ITGoogle Q4_K_M · 10.9 GB | 26.7 tok/s | |
| 4 | Qwen3.8 27BAlibaba Q4_K_M · 17.9 GB | 12.3 tok/s | |
| 5 | Q4_K_M · 6.5 GB | 34.6 tok/s |
The Mac Mini M5 Pro is the fastest 24 GB machine: it runs 14B models well and fits 35B MoE. macOS keeps about 3.5 GB, leaving 20.5 GB for models. With 26 TFLOPS it reads long prompts faster than most 24 GB machines: a 4K-token prompt takes 44 s. M5 Pro generation, released 2026.
Mac Mini M5 Pro strengths and weaknesses
Strengths
- Runs 35B MoE models: Qwen3.6 35B A3B at 49.7 tok/s in 19.8 GB
Weaknesses
- 70B models don't fit: Llama 3.3 70B Instruct needs 51.3 GB at Q4 vs 20.5 GB usable here
- Slow on large dense models: Qwen3.8 27B at 12.3 tok/s
- Slow prompt reading: A 32K-token prompt on Qwen3.8 27B waits 5.9 min for the first token
- No CUDA: vLLM, AWQ, GPTQ and FP8 builds don't run; use GGUF or MLX
Memory for models
20.5 GB of 24 GB70B models need 51.3 GB at Q4, more than the 20.5 GB available. The 48 GB config fits Qwen3.8 27B →
Other memory sizes of the same machine first, then machines and GPUs people weigh against it, each with the one reason it is worth a look. Speeds on Qwen3.5 9B.
Output speed depends on memory bandwidth (307 GB/s). Grey rows need more than the 20.5 GB available.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB33.8tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB22.6tok/s
- 27–31BQwen3.8 27B · Q4_K_M · 17.9 GB12.3tok/s
- 49.7tok/s
- 70BLlama 3.3 70B Instruct · does not fitneeds 51.3 GB
- 120B MoEGPT OSS 120B · does not fitneeds 70.8 GB
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
Pro generations: writing vs compute
M1 Pro → M5 ProWriting speed follows memory bandwidth and barely moved: 22.0 to 33.8 tok/s since the M1 Pro. Compute, which sets how fast prompts are read, grew 2.8×. Indexed to the M1 Pro; speeds on Qwen3.5 9B, same 16 GB.
All 1068 models on the Mac Mini M5 Pro
Featured onlyBest quant that fits in 20.5 GB, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works
| Model | Quant | VRAM | Speed | Context | Score | Fit |
|---|---|---|---|---|---|---|
| Q2_K | 20.2 GB | 13.1 t/s | — | D22, Barely runs | ||
| Q4_K_M | 7.9 GB | 38.8 t/s | 131K | A75, Runs well | ||
| IQ2_M | 18.2 GB | 36.6 t/s | 33K | B61, Decent | ||
Internlm Chat 7BChat | FP16 | 15.4 GB | 14.0 t/s | 2K | B53, Decent | |
XiYanSQL QwenCoder 32B 2504ChatCode | Q3_K_M | 20.2 GB | 13.1 t/s | 33K | D22, Barely runs | |
| IQ2_S | 19.7 GB | 93.2 t/s | — | B62, Decent | ||
Pollux Judge 32BChat | Q2_K | 18.5 GB | 14.6 t/s | 33K | C40, Usable | |
Qwen2.5 Coder 14BChatCode | Q4_K_M | 12.4 GB | 22.5 t/s | 33K | B63, Decent | |
Llama Guard 3 8BChat | BF16 | 20.2 GB | 12.2 t/s | — | D21, Barely runs | |
Yi 34BChat | IQ4_XS | 19.8 GB | 11.1 t/s | 4K | D23, Barely runs | |
Nemotron H 8B Reasoning 128KChatReasoning | BF16 | 20.4 GB | 12.1 t/s | — | D19, Barely runs | |
| Q2_K | 19.4 GB | 44.7 t/s | 262K | B55, Decent | ||
Yi 34B ChatChat | IQ4_XS | 19.8 GB | 11.1 t/s | 4K | D23, Barely runs | |
QwQ 32B PreviewChatReasoning | IQ3_M | 19.3 GB | 13.8 t/s | 33K | C32, Usable | |
C4ai Command R V01ChatFunctions | IQ3_S | 20.2 GB | 13.1 t/s | — | D22, Barely runs |
Specs
- Chip
- M5 Pro
- CPU cores
- 15
- GPU cores
- 16
- Unified memory
- 24 GB
- For models
- 20.5 GB (macOS keeps 3.5 GB)
- Memory bandwidth
- 307 GB/s
- FP16 (dense)
- 25.8 TFLOPS
- Neural Engine
- 38 TOPS
- Released
- 2026-09-22
Mac Mini M5 Pro FAQ
How much memory is available for AI on the Mac Mini M5 Pro (24 GB)?
About 20.5 GB of the 24 GB. macOS keeps roughly 3.5 GB for itself and other apps.
What is the largest model the Mac Mini M5 Pro (24 GB) can run?
Qwen3.6 35B A3B (36.0B) at IQ4_XS uses 19.8 GB, 97% of the available memory, and runs at about 49.7 tok/s.
Can the Mac Mini M5 Pro (24 GB) run Qwen3.8 27B?
Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of the 20.5 GB available, and it is the best model that runs well on this machine.
Can the Mac Mini M5 Pro (24 GB) run 70B models?
No. Llama 3.3 70B Instruct needs 51.3 GB at Q4, more than the 20.5 GB available.
How long does the Mac Mini M5 Pro (24 GB) take to read a long prompt?
On Qwen3.8 27B, about 44 s for a 4K-token prompt and 5.9 min for a 32K-token document before the first word appears.
Is renting a GPU cheaper than buying the Mac Mini M5 Pro?
An RTX 3090 (24 GB) rents from $0.17/hr.