128 GB unified memory and 546 GB/s of memory bandwidth. The strongest single-device fits at 4-bit are Gemma 4 31B (~35 tok/s), Qwen3 32B (~33 tok/s), gpt-oss-120b (~192 tok/s), QwQ 32B (~33 tok/s) and gpt-oss-20b (~260 tok/s).
54 open-weight models fit in MacBook Pro M4 Max 128 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 | ~34.9 | $18.50 | pays back in 254 mo |
| Qwen3 32B Alibaba (Qwen)32.8B | 50.0 | 20 GB | ~33.3 | $15.00 | pays back in 313 mo |
| gpt-oss-120b OpenAI116.8B · 5.7B active | 45.8 | 70 GB | ~192 | $6.75 | pays back in 696 mo |
| QwQ 32B Alibaba (Qwen)32.8B | 37.2 | 20 GB | ~33.3 | — | no API price |
| gpt-oss-20b OpenAI20.9B · 4.2B active | 35.2 | 13 GB | ~260 | $4.05 | pays back in 1160 mo |
| Gemma 3 27B Google27.4B | 28.5 | 16 GB | ~39.9 | $12.00 | pays back in 392 mo |
| Mistral Large 2 Mistral122.6B | 26.8 | 74 GB | ~8.9 | $350 | pays back in 13 mo |
| Llama 3.1 70B Instruct Meta70.6B | 24.3 | 42 GB | ~15.5 | $50.00 | pays back in 94 mo |
| Llama 3.3 70B Instruct Meta70.6B | 20.9 | 42 GB | ~15.5 | $18.00 | pays back in 261 mo |
| Llama 4 Scout Meta108.6B · 17B active | 12.3 | 65 GB | ~64.2 | $14.00 | pays back in 336 mo |
| Nemotron 3.5 Lightning NVIDIA31.6B · 3B active | — | 19 GB | ~364 | $5.50 | pays back in 855 mo |
| Qwen3.6 35B A3B Alibaba (Qwen)36B · 3B active | — | 22 GB | ~364 | $22.50 | pays back in 209 mo |
| Qwen3.5-35B-A3B Alibaba (Qwen)36B · 3B active | — | 22 GB | ~364 | $25.25 | pays back in 186 mo |
| Nemotron 3 Nano 30B A3B NVIDIA31.6B · 3B active | — | 19 GB | ~364 | $10.00 | pays back in 470 mo |
| Qwen3 VL 30B A3B Thinking Alibaba (Qwen)31.1B · 3B active | — | 19 GB | ~364 | $54.00 | pays back in 87 mo |
| Qwen3 VL 30B A3B Instruct Alibaba (Qwen)31.1B · 3B active | — | 19 GB | ~364 | $30.00 | pays back in 157 mo |
| Qwen3 Next 80B A3B Thinking Alibaba (Qwen)81.3B · 3B active | — | 49 GB | ~364 | $45.00 | pays back in 104 mo |
| Qwen3 Next 80B A3B Instruct Alibaba (Qwen)81.3B · 3B active | — | 49 GB | ~364 | $29.20 | pays back in 161 mo |
| Qwen3 Coder 30B A3B Instruct Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~364 | $14.00 | pays back in 336 mo |
| Qwen3 30B A3B Instruct 2507 Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~364 | $16.50 | pays back in 285 mo |
| Hunyuan A13B Instruct Tencent (Hunyuan)80.4B · 3B active | — | 48 GB | ~364 | $28.25 | pays back in 166 mo |
| Qwen3 30B A3B Alibaba (Qwen)30.5B · 3B active | — | 18 GB | ~364 | $24.50 | pays back in 192 mo |
| Ministral 3 3B 2512 Mistral3.8B | — | 2 GB | ~287 | $12.50 | pays back in 376 mo |
| Qwen3 Coder Next Alibaba (Qwen)79.7B · 3.9B active | — | 48 GB | ~280 | $32.00 | pays back in 147 mo |
| Gemma 4 26B A4B Google25.8B · 4B active | — | 16 GB | ~273 | $9.70 | pays back in 485 mo |
| Gemma 3 4B Google4.3B | — | 3 GB | ~254 | $5.25 | pays back in 895 mo |
| GLM 4.7 Flash Zhipu AI (GLM)31.2B · 4.6B active | — | 19 GB | ~237 | $16.05 | pays back in 293 mo |
| Mistral Small 4 Mistral119.4B · 7.1B active | — | 72 GB | ~154 | $30.00 | pays back in 157 mo |
| Stable Diffusion 3.5 Large Stability AI8.1B | — | 5 GB | ~135 | — | no API price |
| Qwen3 VL 8B Instruct Alibaba (Qwen)8.8B | — | 5 GB | ~124 | $43.75 | pays back in 107 mo |
| Ministral 3 8B 2512 Mistral8.9B | — | 5 GB | ~123 | $18.75 | pays back in 251 mo |
| Qwen3.5-9B Alibaba (Qwen)9.7B | — | 6 GB | ~113 | $11.25 | pays back in 418 mo |
| Qwen3.5-122B-A10B Alibaba (Qwen)125.1B · 10B active | — | 75 GB | ~109 | $78.00 | pays back in 60 mo |
| FLUX.1 [dev] Black Forest Labs11.9B | — | 7 GB | ~91.8 | — | no API price |
| Nemotron 3 Super NVIDIA123.6B · 12B active | — | 74 GB | ~91.0 | $18.50 | pays back in 254 mo |
| Gemma 3 12B Google12.2B | — | 7 GB | ~89.5 | $7.00 | pays back in 671 mo |
| Ministral 3 14B 2512 Mistral13.9B | — | 8 GB | ~78.6 | $25.00 | pays back in 188 mo |
| GLM 4.6V Zhipu AI (GLM)107.7B · 14.3B active | — | 65 GB | ~76.4 | $52.50 | pays back in 90 mo |
| GLM 4.5V Zhipu AI (GLM)107.7B · 14.3B active | — | 65 GB | ~76.4 | $105 | pays back in 45 mo |
| Qwen3 14B Alibaba (Qwen)14.8B | — | 9 GB | ~73.8 | $15.50 | pays back in 303 mo |
| GLM 4.5 Air Zhipu AI (GLM)110.5B · 15B active | — | 66 GB | ~72.8 | $34.25 | pays back in 137 mo |
| Mistral Small 3 Mistral23.6B | — | 14 GB | ~46.3 | $7.00 | pays back in 671 mo |
| Mistral Small 3.2 24B Mistral24B | — | 14 GB | ~45.5 | $12.50 | pays back in 376 mo |
| Mistral Small 3.1 24B Mistral24B | — | 14 GB | ~45.5 | $48.97 | pays back in 96 mo |
| Voxtral Small 24B 2507 Mistral24.3B | — | 15 GB | ~44.9 | $17.50 | pays back in 269 mo |
| Qwen3.8 27B Alibaba (Qwen)27.8B | — | 17 GB | ~39.3 | $14.50 | pays back in 324 mo |
| Qwen3.6 27B Alibaba (Qwen)27.8B | — | 17 GB | ~39.3 | $70.00 | pays back in 67 mo |
| Qwen3.5-27B Alibaba (Qwen)27.8B | — | 17 GB | ~39.3 | $58.50 | pays back in 80 mo |
| Muse Glimmer 30B Meta29.8B | — | 18 GB | ~36.6 | $60.00 | pays back in 78 mo |
| Mixtral 8x22B Mistral140.6B · 44B active | — | 84 GB | ~24.8 | $350 | pays back in 13 mo |
| Qwen2.5 72B Instruct Alibaba (Qwen)72.7B | — | 44 GB | ~15.0 | $46.00 | pays back in 102 mo |
| Qwen2.5 VL 72B Instruct Alibaba (Qwen)73.4B | — | 44 GB | ~14.9 | $105 | pays back in 45 mo |
| Command A Cohere111.1B | — | 67 GB | ~9.8 | $500 | pays back in 9 mo |
| Devstral 2 2512 Mistral125B | — | 75 GB | ~8.7 | $90.00 | pays back in 52 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.