Latest (2026)
| Model | Made by | Memory (4-bit, 32K) | Smallest Mac that runs it | Speed there |
|---|---|---|---|---|
| Gemma 4 E4B | 6.9 GB | Mac Mini M6 32GB from $169/mo |
26–36 tokens/sec | |
| IBM Granite Switch 4.1 8B | IBM | 11.8 GB | Mac Mini M6 32GB from $169/mo |
12–16 tokens/sec |
| Gemma 4 12B | 8.8 GB | Mac Mini M6 32GB from $169/mo |
11–15 tokens/sec | |
| Gemma 4 26B-A4B (MoE) | 16.3 GB | Mac Mini M6 32GB from $169/mo |
30–42 tokens/sec | |
| Qwen3.8 27B | Alibaba (Qwen team) | 18.7 GB | Mac Mini M6 32GB from $169/mo |
4.8–6.9 tokens/sec |
| Gemma 4 31B | 21.3 GB | Mac Mini M6 32GB from $169/mo |
4.2–6.1 tokens/sec | |
| Qwen3.6 35B-A3B (MoE) | Alibaba (Qwen team) | 21.8 GB | Mac Mini M6 32GB from $169/mo |
37–51 tokens/sec |
| Flint 40B (MoE, coding) | Recursive Superintelligence | 24.8 GB | Mac Mini M6 32GB from $169/mo |
41–56 tokens/sec |
| NVIDIA Nemotron 3.5 Bolt 74B (MoE) | NVIDIA | 43.6 GB | Mac Mini M5 Pro 64GB from $299/mo |
25–34 tokens/sec |
| Qwen3.8 Flash-Next 180B (MoE, 6B active) | Alibaba (Qwen team) | 104 GB | Mac Studio M4 Max 128GB from $499/mo |
57–77 tokens/sec |
| Xiaomi MiMo-V2.6 Flash 309B (MoE) | Xiaomi | 176 GB | Mac Studio M5 Ultra 256GB from $1,199/mo |
54–73 tokens/sec |
| MiniMax-M3 428B (MoE) | MiniMax | 246 GB | 2× Mac Studio M5 Ultra 256GB Cluster from $2,399/mo |
33–45 tokens/sec |
| Reflection Beam 1.0 501B (MoE) | Reflection AI | 284 GB | 2× Mac Studio M5 Ultra 256GB Cluster from $2,399/mo |
34–46 tokens/sec |
| GLM-5.3 753B (MoE) | Z.ai | 428 GB | 2× Mac Studio M5 Ultra 256GB Cluster from $2,399/mo |
21–29 tokens/sec |
| DeepSeek V4.1 Flash 763B (MoE) | DeepSeek | 433 GB | 2× Mac Studio M5 Ultra 256GB Cluster from $2,399/mo |
45–59 tokens/sec |
| Mistral Large 4 1T (MoE) | Mistral AI | 567 GB | 2× Mac Studio M5 Ultra 512GB Cluster from $4,999/mo |
18–25 tokens/sec |
| Meta Muse Spark Open 816B (MoE) | Meta | 468 GB | 2× Mac Studio M5 Ultra 512GB Cluster from $4,999/mo |
20–27 tokens/sec |
| Qwen3.8 2.4T-A95B (MoE) | Alibaba (Qwen team) | 1.3 TB | 4× Mac Studio M5 Ultra 512GB Cluster from $9,999/mo |
9.4–13 tokens/sec |
Earlier favorites
| Model | Made by | Memory (4-bit, 32K) | Smallest Mac that runs it | Speed there |
|---|---|---|---|---|
| gpt-oss-20b (MoE) | OpenAI | 14.8 GB | Mac Mini M6 32GB from $169/mo |
29–40 tokens/sec |
| gpt-oss-120b (MoE) | OpenAI | 69.6 GB | Mac Studio M4 Max 128GB from $499/mo |
59–79 tokens/sec |
| Llama 3.1 8B | Meta | 10.0 GB | Mac Mini M6 32GB from $169/mo |
13–19 tokens/sec |
| Mistral Small 24B | Mistral AI | 20.0 GB | Mac Mini M6 32GB from $169/mo |
5.1–7.3 tokens/sec |
| Llama 3.3 70B | Meta | 51.2 GB | Mac Mini M5 Pro 64GB from $299/mo |
3.2–4.7 tokens/sec |
| Llama 4 Scout 109B (MoE) | Meta | 68.8 GB | Mac Studio M4 Max 128GB from $499/mo |
21–29 tokens/sec |
| Qwen3 235B-A22B (MoE) | Alibaba (Qwen team) | 140 GB | Mac Studio M5 Ultra 256GB from $1,199/mo |
36–49 tokens/sec |
| Llama 4 Maverick 400B (MoE) | Meta | 233 GB | 2× Mac Studio M5 Ultra 256GB Cluster from $2,399/mo |
40–53 tokens/sec |
| Llama 3.1 405B | Meta | 245 GB | 2× Mac Studio M5 Ultra 256GB Cluster from $2,399/mo |
2.4–3.4 tokens/sec |
| DeepSeek V3 / R1 671B (MoE) | DeepSeek | 381 GB | 2× Mac Studio M5 Ultra 256GB Cluster from $2,399/mo |
23–32 tokens/sec |
| Kimi K2 1T (MoE) | Moonshot AI | 566 GB | 2× Mac Studio M5 Ultra 512GB Cluster from $4,999/mo |
26–35 tokens/sec |
Comparing with GPU clouds? See what the same models cost on rented H100, H200 and B200 GPUs.
Estimates from our AI Model Calculator, using each Mac's published memory bandwidth. Don't see your model? The calculator takes any size.