Apple M4 Max (64GB)
The Apple M4 Max (64GB) uses a unified memory architecture. For reliable local LLM inference without OS memory pressure, this benchmark models a 75% usable allocation (48 GB) with 546 GB/s unified memory bandwidth.
Memory
64 GB
Usable Memory Ceiling
48 GB (75% usable budget)
Bandwidth
546 GB/s
Max Inference Class
32B Models (Q4)
TDP
140W
VRAM Scaling by Context Length
Estimated memory requirements as context length increases. Curves crossing above the reference line indicate Out of Memory (OOM).
Precision:
Best Models to Run on Apple M4 Max (64GB)
Inference throughput, memory compatibility, and generation speeds for open-weights LLMs on Apple M4 Max (64GB).
Sort by:
Context:
KV Cache:
| Rank | Model | Parameters | Precision | TPS (Approx.) | TTFT | Can Run |
|---|---|---|---|---|---|---|
#36 | Qwen 3.8 27B | 27B | INT4 Q8 | 27 tok/s 16 tok/s | ~15.9s ~15.9s | |
#80 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 Q8 | 102 tok/s 74 tok/s | ~3.1s ~3.1s | |
#86 | Qwen3.5-27B | 27B | INT4 Q8 | 27 tok/s 16 tok/s | ~15.9s ~15.9s | |
#93 | Sarvam-30B | 32B (2.4B active) | INT4 Q8 | 128 tok/s 100 tok/s | ~2.2s ~2.2s | |
#99 | Muse Glimmer 30B | 30B | INT4 Q8 | 25 tok/s 14 tok/s | ~17.6s ~17.6s | |
#102 | Phi-4 Reasoning Plus | 14B | INT4 Q8 FP16 | 36 tok/s 24 tok/s 14 tok/s | ~8.4s ~8.4s ~8.4s | |
#103 | Qwen3.6 35B A3B | 35B (3B active) | INT4 Q8 | 117 tok/s 88 tok/s | ~2.5s ~2.5s | |
#105 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 FP16 | 3 tok/s 1 tok/s 1 tok/s | ~173.8s ~174.1s ~174.7s | |
#110 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 Q8 | 108 tok/s 78 tok/s | ~2.8s ~2.8s | |
#117 | Nemotron 3.5 Bolt 74B A9B | 74B | INT4 | 10 tok/s | ~46.8s | |
#120 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 Q8 | 117 tok/s 88 tok/s | ~2.5s ~2.5s | |
#121 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 69 tok/s 43 tok/s 23 tok/s | ~5.4s ~5.4s ~5.4s | |
#127 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 82 tok/s 53 tok/s 29 tok/s | ~4.2s ~4.2s ~4.3s | |
#130 | Gemma 4 12B | 11.95B | INT4 Q8 FP16 | 41 tok/s 28 tok/s 16 tok/s | ~7.2s ~7.2s ~7.2s | |
#136 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 118 tok/s 81 tok/s 48 tok/s | ~2.5s ~2.5s ~2.5s | |
#139 | GPT-OSS 20B | 21B (3.6B active) | INT4 Q8 FP16 | 55 tok/s 48 tok/s 32 tok/s | ~2.7s ~2.7s ~3.1s | |
#140 | GLM-4V | 9B | INT4 Q8 FP16 | 51 tok/s 35 tok/s 21 tok/s | ~5.4s ~5.4s ~5.4s | |
#145 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 24 tok/s 14 tok/s 7 tok/s | ~17.1s ~17.1s ~17.1s | |
#150 | Qwen3-30B-A3B | 30B (3B active) | INT4 Q8 | 118 tok/s 89 tok/s | ~2.5s ~2.5s | |
#152 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 75 tok/s 47 tok/s 26 tok/s | ~4.8s ~4.8s ~4.8s | |
#157 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 75 tok/s 47 tok/s 26 tok/s | ~4.8s ~4.8s ~4.8s | |
#158 | Mistral Large 3 | 41B (675B active) | INT4 | 1 tok/s | ~379.4s | |
#159 | Qwen2.5-72B | 72B | INT4 | 10 tok/s | ~45.6s | |
#161 | GLM-4 | 32B | INT4 Q8 | 18 tok/s 11 tok/s | ~18.8s ~20.4s | |
#162 | DeepSeek-R1 70B | 70B | INT4 | 12 tok/s | ~40.8s | |
#163 | Gemma 3 27B | 27B | INT4 Q8 | 27 tok/s 16 tok/s | ~15.9s ~15.9s | |
#164 | Phi-4 | 14B | INT4 Q8 FP16 | 48 tok/s 29 tok/s 15 tok/s | ~8.4s ~8.4s ~8.4s | |
#165 | Llama 3.3 70B | 70B | INT4 | 11 tok/s | ~40.8s | |
#166 | Qwen3-32B | 32B | INT4 Q8 | 24 tok/s 14 tok/s | ~18.8s ~18.8s | |
#167 | Hunyuan Standard | 52B (389B active) | INT4 | 2 tok/s | ~219.2s | |
#168 | DeepSeek-R1 32B | 32B | INT4 Q8 | 24 tok/s 14 tok/s | ~18.8s ~18.8s | |
#169 | Gemma 4 31B | 30.7B | INT4 Q8 | 24 tok/s 14 tok/s | ~18.0s ~18.0s | |
#171 | Magistral Small | 24B | INT4 Q8 | 23 tok/s 15 tok/s | ~14.2s ~14.2s | |
#172 | Qwen2.5-32B | 32B | INT4 Q8 | 24 tok/s 14 tok/s | ~18.8s ~18.8s | |
#174 | OLMo 3.1 32B Instruct | 32B | INT4 Q8 | 18 tok/s 11 tok/s | ~18.8s ~20.4s | |
#175 | Llama 3.1 70B | 70B | INT4 | 11 tok/s | ~40.8s | |
#176 | DeepSeek-R1 14B | 14B | INT4 Q8 FP16 | 50 tok/s 30 tok/s 16 tok/s | ~8.4s ~8.4s ~8.4s | |
#177 | Ministral 3 14B | 14B | INT4 Q8 FP16 | 36 tok/s 24 tok/s 14 tok/s | ~8.4s ~8.4s ~8.4s | |
#180 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 55 tok/s 33 tok/s 18 tok/s | ~7.2s ~7.2s ~7.2s | |
#183 | Qwen2-72B | 72B | INT4 | 10 tok/s | ~45.6s | |
#184 | Mistral-Small-2501 | 24B | INT4 Q8 FP16 | 30 tok/s 18 tok/s 8 tok/s | ~14.2s ~14.2s ~16.7s | |
#185 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 55 tok/s 39 tok/s 23 tok/s | ~4.8s ~4.8s ~4.8s | |
#186 | Gemma 2 27B | 27B | INT4 Q8 | 27 tok/s 16 tok/s | ~15.9s ~15.9s | |
#187 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 61 tok/s 43 tok/s 26 tok/s | ~4.2s ~4.2s ~4.3s | |
#188 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 166 tok/s 125 tok/s 81 tok/s | ~1.3s ~1.3s ~1.3s | |
#189 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 166 tok/s 125 tok/s 81 tok/s | ~1.3s ~1.3s ~1.3s | |
#190 | Qwen2.5-14B | 14B | INT4 Q8 FP16 | 48 tok/s 29 tok/s 15 tok/s | ~8.4s ~8.4s ~8.4s | |
#191 | OLMo 3.1 32B Think | 32B | INT4 Q8 | 18 tok/s 11 tok/s | ~18.8s ~20.4s | |
#192 | Llama 3 70B | 70B | INT4 | 11 tok/s | ~40.8s | |
#193 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 102 tok/s 68 tok/s 39 tok/s | ~3.1s ~3.1s ~3.1s | |
#195 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 97 tok/s 77 tok/s 52 tok/s | ~1.9s ~1.9s ~1.9s | |
#196 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 77 tok/s 48 tok/s 26 tok/s | ~4.8s ~4.8s ~4.8s | |
#198 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 69 tok/s 43 tok/s 23 tok/s | ~5.4s ~5.4s ~5.4s | |
#199 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 Q8 | 118 tok/s 89 tok/s | ~2.5s ~2.5s | |
#200 | Kimi Linear 48B A3B Instruct | 48B (3B active) | INT4 | 43 tok/s | ~2.7s | |
#201 | Qwen3-4B | 4B | INT4 Q8 FP16 | 118 tok/s 81 tok/s 48 tok/s | ~2.5s ~2.5s ~2.5s | |
#202 | Qwen3-14B | 14B | INT4 Q8 FP16 | 48 tok/s 29 tok/s 15 tok/s | ~8.4s ~8.4s ~8.4s | |
#203 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 118 tok/s 81 tok/s 48 tok/s | ~2.5s ~2.5s ~2.5s | |
#204 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 84 tok/s 53 tok/s 30 tok/s | ~4.2s ~4.2s ~4.3s | |
#205 | OLMo 3 32B Base | 32B | INT4 Q8 | 18 tok/s 11 tok/s | ~18.8s ~20.4s | |
#206 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 227 tok/s 193 tok/s 145 tok/s | ~537 ms ~538 ms ~539 ms | |
#207 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 75 tok/s 47 tok/s 26 tok/s | ~4.8s ~4.8s ~4.8s | |
#207 | Qwen3-8B | 8B | INT4 Q8 FP16 | 75 tok/s 47 tok/s 26 tok/s | ~4.8s ~4.8s ~4.8s | |
#210 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 Q8 | 50 tok/s 46 tok/s | ~2.5s ~2.5s | |
#211 | Command R | 35B | INT4 Q8 | 17 tok/s 10 tok/s | ~20.5s ~22.3s | |
#212 | Llama 3 8B | 8B | INT4 Q8 FP16 | 75 tok/s 47 tok/s 26 tok/s | ~4.8s ~4.8s ~4.8s | |
#213 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 177 tok/s 137 tok/s 91 tok/s | ~1.1s ~1.1s ~1.1s | |
#214 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 254 tok/s 239 tok/s 210 tok/s | ~179 ms ~179 ms ~180 ms | |
#216 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 243 tok/s 213 tok/s 167 tok/s | ~397 ms ~398 ms ~399 ms | |
#217 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 122 tok/s 84 tok/s 50 tok/s | ~2.4s ~2.4s ~2.4s | |
#218 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 82 tok/s 53 tok/s 29 tok/s | ~4.2s ~4.2s ~4.3s | |
#219 | Mixtral-8x7B-v0.1 | 46.7B (7B active) | INT4 Q8 | 74 tok/s 44 tok/s | ~4.9s ~5.8s | |
#220 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 75 tok/s 47 tok/s 26 tok/s | ~4.8s ~4.8s ~4.8s | |
#221 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 186 tok/s 147 tok/s 100 tok/s | ~997 ms ~999 ms ~1.0s | |
#222 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 138 tok/s 98 tok/s 60 tok/s | ~1.9s ~1.9s ~1.9s | |
#223 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 122 tok/s 84 tok/s 50 tok/s | ~2.4s ~2.4s ~2.4s | |
#224 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 97 tok/s 84 tok/s 62 tok/s | ~1.4s ~1.4s ~1.4s | |
#225 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 166 tok/s 125 tok/s 81 tok/s | ~1.3s ~1.3s ~1.3s | |
#226 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 80 tok/s 51 tok/s 28 tok/s | ~4.4s ~4.4s ~4.4s | |
#227 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 212 tok/s 176 tok/s 128 tok/s | ~682 ms ~683 ms ~685 ms | |
#228 | Phi-3-medium | 14B | INT4 Q8 FP16 | 48 tok/s 29 tok/s 15 tok/s | ~8.4s ~8.4s ~8.4s | |
#229 | Yi-34B | 34B | INT4 Q8 | 18 tok/s 11 tok/s | ~19.9s ~21.7s | |
#230 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 61 tok/s 43 tok/s 26 tok/s | ~4.2s ~4.2s ~4.3s | |
#231 | Qwen2-7B | 7B | INT4 Q8 FP16 | 82 tok/s 53 tok/s 29 tok/s | ~4.2s ~4.2s ~4.3s | |
#232 | Phi-3-small | 7B | INT4 Q8 FP16 | 82 tok/s 53 tok/s 29 tok/s | ~4.2s ~4.2s ~4.3s | |
#233 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 80 tok/s 51 tok/s 28 tok/s | ~4.4s ~4.4s ~4.4s | |
#234 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 212 tok/s 176 tok/s 128 tok/s | ~682 ms ~683 ms ~685 ms | |
#236 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 61 tok/s 43 tok/s 26 tok/s | ~4.2s ~4.2s ~4.3s | |
#237 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 174 tok/s 130 tok/s 83 tok/s | ~1.3s ~1.3s ~1.3s | |
#238 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 67 tok/s 49 tok/s 30 tok/s | ~3.7s ~3.7s ~3.7s | |
#239 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 67 tok/s 49 tok/s 30 tok/s | ~3.7s ~3.7s ~3.7s | |
#240 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 67 tok/s 49 tok/s 30 tok/s | ~3.7s ~3.7s ~3.7s | |
Unranked Models (Sorted by Speed) | ||||||
- | DINOv3 ViT-L/16 | 0.3B | INT4 Q8 FP16 | 270 tok/s 251 tok/s 216 tok/s | ~199 ms ~199 ms ~200 ms | |
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 252 tok/s 224 tok/s 181 tok/s | ~330 ms ~330 ms ~331 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 252 tok/s 224 tok/s 181 tok/s | ~330 ms ~330 ms ~331 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 248 tok/s 233 tok/s 203 tok/s | ~199 ms ~199 ms ~200 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 248 tok/s 233 tok/s 203 tok/s | ~199 ms ~199 ms ~200 ms | |
- | Liquid D1-Omni 600M | 0.6B | INT4 Q8 FP16 | 243 tok/s 213 tok/s 167 tok/s | ~397 ms ~398 ms ~399 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 221 tok/s 182 tok/s 131 tok/s | ~682 ms ~683 ms ~685 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 212 tok/s 176 tok/s 128 tok/s | ~682 ms ~683 ms ~685 ms | |
- | MiniCPM-V 4.7 1B | 1B | INT4 Q8 FP16 | 212 tok/s 176 tok/s 128 tok/s | ~682 ms ~683 ms ~685 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 186 tok/s 147 tok/s 100 tok/s | ~997 ms ~999 ms ~1.0s | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 186 tok/s 147 tok/s 100 tok/s | ~997 ms ~999 ms ~1.0s | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 177 tok/s 137 tok/s 91 tok/s | ~1.1s ~1.1s ~1.1s | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 174 tok/s 133 tok/s 88 tok/s | ~1.2s ~1.2s ~1.2s | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 145 tok/s 101 tok/s 61 tok/s | ~1.9s ~1.9s ~1.9s | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 142 tok/s 123 tok/s 92 tok/s | ~870 ms ~872 ms ~874 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 142 tok/s 123 tok/s 92 tok/s | ~870 ms ~872 ms ~874 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 142 tok/s 123 tok/s 92 tok/s | ~870 ms ~872 ms ~874 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 140 tok/s 99 tok/s 60 tok/s | ~1.9s ~1.9s ~1.9s | |
- | Mellum 2.1 12B A2.5B Thinking | 12B (2.5B active) | INT4 Q8 FP16 | 140 tok/s 105 tok/s 67 tok/s | ~1.9s ~1.9s ~1.9s | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 138 tok/s 98 tok/s 60 tok/s | ~1.9s ~1.9s ~1.9s | |
- | LFM2-VL 3B | 3B | INT4 Q8 FP16 | 138 tok/s 98 tok/s 60 tok/s | ~1.9s ~1.9s ~1.9s | |
- | Liquid D1 3B | 3B | INT4 Q8 FP16 | 138 tok/s 98 tok/s 60 tok/s | ~1.9s ~1.9s ~1.9s | |
- | Liquid D1-VL 3B | 3B | INT4 Q8 FP16 | 138 tok/s 98 tok/s 60 tok/s | ~1.9s ~1.9s ~1.9s | |
- | Liquid LFM2-VL 3B D1 | 3B | INT4 Q8 FP16 | 138 tok/s 98 tok/s 60 tok/s | ~1.9s ~1.9s ~1.9s | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 138 tok/s 98 tok/s 60 tok/s | ~1.9s ~1.9s ~1.9s | |
- | LFM2.5 24B A2B | 24B (2.5B active) | INT4 Q8 FP16 | 130 tok/s 100 tok/s 58 tok/s | ~2.1s ~2.1s ~2.5s | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 FP16 | 123 tok/s 91 tok/s 54 tok/s | ~2.3s ~2.3s ~2.5s | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 FP16 | 123 tok/s 91 tok/s 54 tok/s | ~2.3s ~2.3s ~2.5s | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 Q8 | 119 tok/s 89 tok/s | ~2.4s ~2.4s | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 Q8 | 119 tok/s 89 tok/s | ~2.4s ~2.4s | |
- | FrogNano 4B | 4B | INT4 Q8 FP16 | 118 tok/s 81 tok/s 48 tok/s | ~2.5s ~2.5s ~2.5s | |
- | Hy-Memory 30B A3B | 30B (3B active) | INT4 Q8 | 118 tok/s 89 tok/s | ~2.5s ~2.5s | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 Q8 | 118 tok/s 89 tok/s | ~2.5s ~2.5s | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 118 tok/s 81 tok/s 48 tok/s | ~2.5s ~2.5s ~2.5s | |
- | MiniCPM-V 4.7 35B A3B | 35B (3B active) | INT4 Q8 | 117 tok/s 88 tok/s | ~2.5s ~2.5s | |
- | EmbeddingGemma 2 | 2B | INT4 Q8 FP16 | 114 tok/s 95 tok/s 68 tok/s | ~1.3s ~1.3s ~1.3s | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 102 tok/s 82 tok/s 55 tok/s | ~1.7s ~1.7s ~1.7s | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 97 tok/s 77 tok/s 52 tok/s | ~1.9s ~1.9s ~1.9s | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 97 tok/s 77 tok/s 52 tok/s | ~1.9s ~1.9s ~1.9s | |
- | AliceAI Foundation 80B A3B Base | 80B (3B active) | INT4 | 91 tok/s | ~3.7s | |
- | Kolibri 1 | 78B (3.5B active) | INT4 | 87 tok/s | ~4.0s | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 86 tok/s 54 tok/s 30 tok/s | ~4.2s ~4.2s ~4.3s | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 82 tok/s 53 tok/s 29 tok/s | ~4.2s ~4.2s ~4.3s | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 82 tok/s 53 tok/s 29 tok/s | ~4.2s ~4.2s ~4.3s | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 80 tok/s 51 tok/s 28 tok/s | ~4.4s ~4.4s ~4.4s | |
- | Youtu Parsing Omni | 8B | INT4 Q8 FP16 | 77 tok/s 48 tok/s 26 tok/s | ~4.8s ~4.8s ~4.8s | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 69 tok/s 43 tok/s 23 tok/s | ~5.4s ~5.4s ~5.4s | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 69 tok/s 43 tok/s 23 tok/s | ~5.4s ~5.4s ~5.4s | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 67 tok/s 49 tok/s 30 tok/s | ~3.7s ~3.7s ~3.7s | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 67 tok/s 49 tok/s 30 tok/s | ~3.7s ~3.7s ~3.7s | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 63 tok/s 39 tok/s 21 tok/s | ~6.0s ~6.0s ~6.0s | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 FP16 | 63 tok/s 56 tok/s 41 tok/s | ~2.2s ~2.2s ~2.2s | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 61 tok/s 37 tok/s 20 tok/s | ~6.6s ~6.6s ~6.6s | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 61 tok/s 43 tok/s 26 tok/s | ~4.2s ~4.2s ~4.3s | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 61 tok/s 43 tok/s 26 tok/s | ~4.2s ~4.2s ~4.3s | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 60 tok/s 43 tok/s 26 tok/s | ~4.3s ~4.3s ~4.3s | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 60 tok/s 43 tok/s 26 tok/s | ~4.3s ~4.3s ~4.3s | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 55 tok/s 39 tok/s 23 tok/s | ~4.8s ~4.8s ~4.8s | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 55 tok/s 39 tok/s 23 tok/s | ~4.8s ~4.8s ~4.8s | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 51 tok/s 35 tok/s 21 tok/s | ~5.4s ~5.4s ~5.4s | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 51 tok/s 35 tok/s 21 tok/s | ~5.4s ~5.4s ~5.4s | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 51 tok/s 35 tok/s 21 tok/s | ~5.4s ~5.4s ~5.4s | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 51 tok/s 35 tok/s 21 tok/s | ~5.4s ~5.4s ~5.4s | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 50 tok/s 35 tok/s 21 tok/s | ~5.5s ~5.5s ~5.6s | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 50 tok/s 35 tok/s 21 tok/s | ~5.5s ~5.5s ~5.6s | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 FP16 | 36 tok/s 24 tok/s 14 tok/s | ~8.4s ~8.4s ~8.4s | |
- | AREX-2 | 27B | INT4 Q8 | 27 tok/s 16 tok/s | ~15.9s ~15.9s | |
- | PPLX Decider v1 27B | 27B | INT4 Q8 | 27 tok/s 16 tok/s | ~15.9s ~15.9s | |
- | PPLX Decider v1.1 27B | 27B | INT4 Q8 | 27 tok/s 16 tok/s | ~15.9s ~15.9s | |
- | Agnes 3.0 Flash | 33B | INT4 Q8 | 23 tok/s 13 tok/s | ~19.3s ~19.4s | |
- | Falcon-40B | 40B | INT4 Q8 | 20 tok/s 11 tok/s | ~23.3s ~25.4s | |
- | Kimi-Dev-72B | 72B | INT4 | 8 tok/s | ~49.3s | |
- | Typhoon-2-70B | 70B | INT4 | 8 tok/s | ~48.0s | |
Inference Capacity
Maximum model parameter class runnable by weight precision.
Context:
4-bit
32B Models
8-bit
32B Models
16-bit
14B Models
Fine-Tuning Capacity
Hardware capacity for local training and adapter fine-tuning.
Context Length:
QLoRA
32B Models
LoRA
14B Models
Full Parameter Training
7B-8B Models
Workload Recommendations
Guidance for optimal precision and operational ceilings on Apple M4 Max (64GB).
Inference Sweet Spot
Optimized for 14B models at uncompressed or Q8 precision, and 32B models at Q4 with up to 32k context.
Bandwidth & Speed Profile
With 546 GB/s aggregate bandwidth, batch size 1 inference operates in a memory-bandwidth bound regime. Generates approximately 121 tok/s on an 8B Q4 model and 14 tok/s on a 70B Q4 model.
Unified Memory Budget
Modeled with a 75% usable allocation budget to preserve system RAM for macOS and display compositor processes.
Frequently Asked Questions
©2025 ApX Machine Learning