48 GB unified memory and 273 GB/s of memory bandwidth. The strongest single-device fits at 4-bit are Gemma 4 31B (~17 tok/s), Qwen3 32B (~17 tok/s), QwQ 32B (~17 tok/s), gpt-oss-20b (~130 tok/s) and Gemma 3 27B (~20 tok/s).
34 open-weight models fit in Mac mini M4 Pro 48 GB at 4-bit, ranked by Analyzer Index.
| Model | Index | Memory | Est. tok/s | API $/month | Verdict |
|---|---|---|---|---|---|
| Gemma 4 31B Google31.3B | 57.4 | 19 GB | ~17.4 | $18.50 | pays back in 97 mo |
| Qwen3 32B Alibaba (Qwen)32.8B | 50.0 | 20 GB | ~16.6 | $15.00 | pays back in 120 mo |
| QwQ 32B Alibaba (Qwen)32.8B | 37.2 | 20 GB | ~16.6 | — | no API price |
| gpt-oss-20b OpenAI20.9B · 4.2B active | 35.2 | 13 GB | ~130 | $4.05 | pays back in 444 mo |
| Gemma 3 27B Google27.4B | 28.5 | 16 GB | ~19.9 | $12.00 | pays back in 150 mo |
| Nemotron 3.5 Lightning NVIDIA31.6B · 3B active | — | 19 GB | ~182 | $5.50 | pays back in 327 mo |
| Qwen3.6 35B A3B Alibaba (Qwen)36B · 3B active | — | 22 GB | ~182 | $22.50 | pays back in 80 mo |
| Qwen3.5-35B-A3B Alibaba (Qwen)36B · 3B active | — | 22 GB | ~182 | $25.25 | pays back in 71 mo |
| Nemotron 3 Nano 30B A3B NVIDIA31.6B · 3B active | — | 19 GB | ~182 | $10.00 | pays back in 180 mo |
| Qwen3 VL 30B A3B Thinking Alibaba (Qwen)31.1B · 3B active | — | 19 GB | ~182 | $54.00 | pays back in 33 mo |
| Qwen3 VL 30B A3B Instruct Alibaba (Qwen)31.1B · 3B active | — | 19 GB | ~182 | $30.00 | pays back in 60 mo |
| Qwen3 Coder 30B A3B Instruct Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~182 | $14.00 | pays back in 129 mo |
| Qwen3 30B A3B Instruct 2507 Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~182 | $16.50 | pays back in 109 mo |
| Qwen3 30B A3B Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~182 | $24.50 | pays back in 73 mo |
| Ministral 3 3B 2512 Mistral3.8B | — | 2 GB | ~144 | $12.50 | pays back in 144 mo |
| Gemma 4 26B A4B Google25.8B · 4B active | — | 16 GB | ~137 | $9.70 | pays back in 186 mo |
| Gemma 3 4B Google4.3B | — | 3 GB | ~127 | $5.25 | pays back in 343 mo |
| GLM 4.7 Flash Zhipu AI (GLM)31.2B · 4.6B active | — | 19 GB | ~119 | $16.05 | pays back in 112 mo |
| Stable Diffusion 3.5 Large Stability AI8.1B | — | 5 GB | ~67.4 | — | no API price |
| Qwen3 VL 8B Instruct Alibaba (Qwen)8.8B | — | 5 GB | ~62.0 | $43.75 | pays back in 41 mo |
| Ministral 3 8B 2512 Mistral8.9B | — | 5 GB | ~61.3 | $18.75 | pays back in 96 mo |
| Qwen3.5-9B Alibaba (Qwen)9.7B | — | 6 GB | ~56.3 | $11.25 | pays back in 160 mo |
| FLUX.1 [dev] Black Forest Labs11.9B | — | 7 GB | ~45.9 | — | no API price |
| Gemma 3 12B Google12.2B | — | 7 GB | ~44.8 | $7.00 | pays back in 257 mo |
| Ministral 3 14B 2512 Mistral13.9B | — | 8 GB | ~39.3 | $25.00 | pays back in 72 mo |
| Qwen3 14B Alibaba (Qwen)14.8B | — | 9 GB | ~36.9 | $15.50 | pays back in 116 mo |
| Mistral Small 3 Mistral23.6B | — | 14 GB | ~23.1 | $7.00 | pays back in 257 mo |
| Mistral Small 3.2 24B Mistral24B | — | 14 GB | ~22.8 | $12.50 | pays back in 144 mo |
| Mistral Small 3.1 24B Mistral24B | — | 14 GB | ~22.8 | $48.97 | pays back in 37 mo |
| Voxtral Small 24B 2507 Mistral24.3B | — | 15 GB | ~22.5 | $17.50 | pays back in 103 mo |
| Qwen3.8 27B Alibaba (Qwen)27.8B | — | 17 GB | ~19.6 | $14.50 | pays back in 124 mo |
| Qwen3.6 27B Alibaba (Qwen)27.8B | — | 17 GB | ~19.6 | $70.00 | pays back in 26 mo |
| Qwen3.5-27B Alibaba (Qwen)27.8B | — | 17 GB | ~19.6 | $58.50 | pays back in 31 mo |
| Muse Glimmer 30B Meta29.8B | — | 18 GB | ~18.3 | $60.00 | pays back in 30 mo |
Memory need is weights × 1.2 for runtime overhead. Speed is a theoretical single-stream ceiling (memory bandwidth ÷ active-weight bytes); real engines usually reach 50–80% of it, and batching many requests raises total throughput well beyond one stream. Rental and purchase prices are typical figures checked 2026-10-14; power and engineering time are not included.