趋近智
Apple M4 (32GB)
Apple M4 (32GB) 采用统一内存架构。为了在避免系统内存压力的情况下确保本地大模型稳定推理，本评测以 75% 可用配额（24 GB）及 120 GB/s 统一内存带宽进行建模。
总显存
32 GB
可用显存上限
24 GB （75% 可用预算）
显存带宽
120 GB/s
最大推理模型级别
32B 模型 (Q4)
热设计功耗
45W
显存占用随上下文长度变化曲线
随上下文长度增长的显存需求评估。当曲线超过参考线时将导致显存不足（OOM）。
量化精度:
Apple M4 (32GB) 最佳运行大模型与速度矩阵
主流开源大模型在 Apple M4 (32GB) 上的推理速度、显存兼容性与生成吞吐表现。
排序:
上下文长度:
KV 缓存:
| 排名 | 模型 | 参数量 | 量化精度 | 预估 TPS | TTFT | 可运行 |
|---|---|---|---|---|---|---|
#48 | Qwen 3.8 27B | 27B | INT4 | 6 tok/s | ~63.6s | |
#83 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 | 29 tok/s | ~13.5s | |
#93 | Qwen3.5-27B | 27B | INT4 | 6 tok/s | ~63.6s | |
#105 | Sarvam-30B | 32B (2.4B active) | INT4 | 42 tok/s | ~8.7s | |
#106 | Qwen3.6 35B A3B | 35B (3B active) | INT4 | 35 tok/s | ~11.0s | |
#110 | Muse Glimmer 30B | 30B | INT4 | 6 tok/s | ~70.6s | |
#111 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 | 1 tok/s 1 tok/s | ~696.5s ~758.4s | |
#112 | Phi-4 Reasoning Plus | 14B | INT4 Q8 | 9 tok/s 6 tok/s | ~33.5s ~33.5s | |
#117 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 | 33 tok/s | ~11.4s | |
#125 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 18 tok/s 11 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
#126 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 | 35 tok/s | ~11.0s | |
#128 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 23 tok/s 13 tok/s 7 tok/s | ~16.9s ~17.0s ~17.0s | |
#135 | Gemma 4 12B | 11.95B | INT4 Q8 | 10 tok/s 7 tok/s | ~28.7s ~28.7s | |
#140 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 37 tok/s 22 tok/s 12 tok/s | ~9.9s ~9.9s ~10.0s | |
#143 | GPT-OSS 20B | 21B (3.6B active) | INT4 | 14 tok/s | ~10.7s | |
#144 | GLM-4V | 9B | INT4 Q8 FP16 | 13 tok/s 9 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
#150 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 6 tok/s 3 tok/s 2 tok/s | ~68.3s ~68.4s ~68.7s | |
#153 | Qwen3-30B-A3B | 30B (3B active) | INT4 | 37 tok/s | ~9.9s | |
#158 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 20 tok/s 12 tok/s 6 tok/s | ~19.3s ~19.3s ~19.4s | |
#159 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 20 tok/s 12 tok/s 6 tok/s | ~19.3s ~19.3s ~19.4s | |
#165 | GLM-4 | 32B | INT4 | 4 tok/s | ~88.5s | |
#166 | Phi-4 | 14B | INT4 Q8 | 12 tok/s 7 tok/s | ~33.5s ~33.5s | |
#167 | Gemma 3 27B | 27B | INT4 | 6 tok/s | ~63.6s | |
#170 | DeepSeek-R1 32B | 32B | INT4 | 6 tok/s | ~75.2s | |
#171 | Qwen3-32B | 32B | INT4 | 6 tok/s | ~75.2s | |
#175 | Magistral Small | 24B | INT4 | 5 tok/s | ~56.8s | |
#176 | OLMo 3.1 32B Instruct | 32B | INT4 | 4 tok/s | ~88.5s | |
#177 | Qwen2.5-32B | 32B | INT4 | 6 tok/s | ~75.2s | |
#178 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 16 tok/s 11 tok/s 6 tok/s | ~17.0s ~17.0s ~17.1s | |
#179 | DeepSeek-R1 14B | 14B | INT4 Q8 | 13 tok/s 7 tok/s | ~33.5s ~33.5s | |
#183 | Ministral 3 14B | 14B | INT4 Q8 | 9 tok/s 6 tok/s | ~33.5s ~33.5s | |
#186 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 14 tok/s 8 tok/s 4 tok/s | ~28.8s ~28.8s ~34.0s | |
#188 | Mistral-Small-2501 | 24B | INT4 | 7 tok/s | ~56.7s | |
#189 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 14 tok/s 10 tok/s 5 tok/s | ~19.3s ~19.3s ~19.4s | |
#190 | Gemma 4 31B | 30.7B | INT4 | 6 tok/s | ~72.2s | |
#191 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
#192 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
#193 | Gemma 2 27B | 27B | INT4 | 6 tok/s | ~63.6s | |
#194 | Qwen2.5-14B | 14B | INT4 Q8 | 12 tok/s 7 tok/s | ~33.5s ~33.5s | |
#195 | OLMo 3.1 32B Think | 32B | INT4 | 4 tok/s | ~88.5s | |
#196 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 | 12 tok/s | ~10.8s | |
#200 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 30 tok/s 18 tok/s 9 tok/s | ~12.5s ~12.5s ~12.6s | |
#202 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 21 tok/s 12 tok/s 6 tok/s | ~19.3s ~19.3s ~19.4s | |
#204 | Qwen3-4B | 4B | INT4 Q8 FP16 | 37 tok/s 22 tok/s 12 tok/s | ~9.9s ~9.9s ~10.0s | |
#205 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 29 tok/s 21 tok/s 13 tok/s | ~7.6s ~7.6s ~7.6s | |
#206 | OLMo 3 32B Base | 32B | INT4 | 4 tok/s | ~88.5s | |
#207 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 18 tok/s 11 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
#208 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 120 tok/s 84 tok/s 51 tok/s | ~2.1s ~2.1s ~2.1s | |
#209 | Qwen3-14B | 14B | INT4 Q8 | 12 tok/s 7 tok/s | ~33.5s ~33.5s | |
#210 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 37 tok/s 22 tok/s 12 tok/s | ~9.9s ~9.9s ~10.0s | |
#211 | Qwen3-8B | 8B | INT4 Q8 FP16 | 20 tok/s 12 tok/s 6 tok/s | ~19.3s ~19.3s ~19.4s | |
#213 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 24 tok/s 14 tok/s 7 tok/s | ~16.9s ~17.0s ~17.0s | |
#214 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 72 tok/s 47 tok/s 26 tok/s | ~4.5s ~4.5s ~4.5s | |
#215 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 20 tok/s 12 tok/s 6 tok/s | ~19.3s ~19.3s ~19.4s | |
#216 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 161 tok/s 137 tok/s 101 tok/s | ~693 ms ~694 ms ~697 ms | |
#217 | Command R | 35B | INT4 | 4 tok/s | ~96.5s | |
#218 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 143 tok/s 104 tok/s 64 tok/s | ~1.6s ~1.6s ~1.6s | |
#219 | Llama 3 8B | 8B | INT4 Q8 FP16 | 20 tok/s 12 tok/s 6 tok/s | ~19.3s ~19.3s ~19.4s | |
#221 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 39 tok/s 24 tok/s 13 tok/s | ~9.5s ~9.5s ~9.5s | |
#223 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 23 tok/s 13 tok/s 7 tok/s | ~16.9s ~17.0s ~17.0s | |
#224 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 20 tok/s 12 tok/s 6 tok/s | ~19.3s ~19.3s ~19.4s | |
#225 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 79 tok/s 52 tok/s 29 tok/s | ~4.0s ~4.0s ~4.0s | |
#226 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 47 tok/s 29 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
#227 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 | 37 tok/s | ~9.9s | |
#228 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 39 tok/s 24 tok/s 13 tok/s | ~9.5s ~9.5s ~9.5s | |
#229 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 29 tok/s 24 tok/s 16 tok/s | ~5.6s ~5.6s ~5.7s | |
#230 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~17.6s ~17.7s ~17.7s | |
#231 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 103 tok/s 71 tok/s 42 tok/s | ~2.7s ~2.7s ~2.7s | |
#232 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
#233 | Phi-3-medium | 14B | INT4 Q8 | 12 tok/s 7 tok/s | ~33.5s ~33.5s | |
#234 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 16 tok/s 11 tok/s 6 tok/s | ~17.0s ~17.0s ~17.1s | |
#235 | Yi-34B | 34B | INT4 | 4 tok/s | ~93.9s | |
#236 | Qwen2-7B | 7B | INT4 Q8 FP16 | 23 tok/s 13 tok/s 7 tok/s | ~16.9s ~17.0s ~17.0s | |
#237 | Phi-3-small | 7B | INT4 Q8 FP16 | 23 tok/s 13 tok/s 7 tok/s | ~16.9s ~17.0s ~17.0s | |
#238 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 103 tok/s 71 tok/s 42 tok/s | ~2.7s ~2.7s ~2.7s | |
#239 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~17.6s ~17.7s ~17.7s | |
#241 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 16 tok/s 11 tok/s 6 tok/s | ~17.0s ~17.0s ~17.1s | |
#242 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 70 tok/s 43 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
#243 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 18 tok/s 12 tok/s 7 tok/s | ~14.6s ~14.6s ~14.7s | |
#244 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 18 tok/s 12 tok/s 7 tok/s | ~14.6s ~14.6s ~14.7s | |
#245 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 18 tok/s 12 tok/s 7 tok/s | ~14.6s ~14.6s ~14.7s | |
未上榜模型（按推理速度排序） | ||||||
- | DINOv3 ViT-L/16 | 0.3B | INT4 Q8 FP16 | 196 tok/s 155 tok/s 107 tok/s | ~772 ms ~773 ms ~776 ms | |
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 157 tok/s 117 tok/s 74 tok/s | ~1.3s ~1.3s ~1.3s | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 157 tok/s 117 tok/s 74 tok/s | ~1.3s ~1.3s ~1.3s | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 152 tok/s 128 tok/s 93 tok/s | ~773 ms ~775 ms ~778 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 152 tok/s 128 tok/s 93 tok/s | ~773 ms ~775 ms ~778 ms | |
- | Liquid D1-Omni 600M | 0.6B | INT4 Q8 FP16 | 143 tok/s 104 tok/s 64 tok/s | ~1.6s ~1.6s ~1.6s | |
- | PixArt-α XL 2 1024-MS | 0.61B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~1.6s ~1.6s ~1.6s | |
- | PixArt-Σ XL 2 1024-MS | 0.61B | INT4 Q8 FP16 | 142 tok/s 103 tok/s 64 tok/s | ~1.6s ~1.6s ~1.6s | |
- | Nemotron 3.5 Parse | 0.63B | INT4 Q8 FP16 | 139 tok/s 100 tok/s 62 tok/s | ~1.7s ~1.7s ~1.7s | |
- | SD 2.1 Turbo | 0.86B | INT4 Q8 FP16 | 115 tok/s 80 tok/s 48 tok/s | ~2.3s ~2.3s ~2.3s | |
- | Stable Diffusion 1.4 | 0.86B | INT4 Q8 FP16 | 115 tok/s 80 tok/s 48 tok/s | ~2.3s ~2.3s ~2.3s | |
- | Stable Diffusion 2.1 | 0.86B | INT4 Q8 FP16 | 115 tok/s 80 tok/s 48 tok/s | ~2.3s ~2.3s ~2.3s | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 113 tok/s 75 tok/s 43 tok/s | ~2.7s ~2.7s ~2.7s | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 103 tok/s 71 tok/s 42 tok/s | ~2.7s ~2.7s ~2.7s | |
- | MiniCPM-V 4.7 1B | 1B | INT4 Q8 FP16 | 103 tok/s 71 tok/s 42 tok/s | ~2.7s ~2.7s ~2.7s | |
- | Kandinsky 2.2 Decoder | 1.22B | INT4 Q8 FP16 | 91 tok/s 61 tok/s 35 tok/s | ~3.3s ~3.3s ~3.3s | |
- | Wan 2.1 T2V 1.3B | 1.3B | INT4 Q8 FP16 | 87 tok/s 58 tok/s 33 tok/s | ~3.5s ~3.5s ~3.5s | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 79 tok/s 52 tok/s 29 tok/s | ~4.0s ~4.0s ~4.0s | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 79 tok/s 52 tok/s 29 tok/s | ~4.0s ~4.0s ~4.0s | |
- | SVD Image-to-Video XT | 1.5B | INT4 Q8 FP16 | 79 tok/s 52 tok/s 29 tok/s | ~4.0s ~4.0s ~4.0s | |
- | SVD Image-to-Video XT 1.1 | 1.5B | INT4 Q8 FP16 | 79 tok/s 52 tok/s 29 tok/s | ~4.0s ~4.0s ~4.0s | |
- | LFM3 1.7B | 1.7B | INT4 Q8 FP16 | 72 tok/s 47 tok/s 26 tok/s | ~4.5s ~4.5s ~4.5s | |
- | LFM3 1.7B Base | 1.7B | INT4 Q8 FP16 | 72 tok/s 47 tok/s 26 tok/s | ~4.5s ~4.5s ~4.5s | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 72 tok/s 47 tok/s 26 tok/s | ~4.5s ~4.5s ~4.5s | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 69 tok/s 45 tok/s 25 tok/s | ~4.7s ~4.7s ~4.8s | |
- | CogVideoX 2B | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
- | LFM3 Audio 2B | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
- | LFM3 Omni 2B | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
- | LFM3 VL 2B | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
- | Liquid D1 2B | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
- | LTX-Video 0.9.1 | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
- | Moondream 3.1 | 9B (2B active) | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.9s ~5.9s ~6.0s | |
- | Stable Diffusion 3 Medium | 2B | INT4 Q8 FP16 | 64 tok/s 41 tok/s 23 tok/s | ~5.3s ~5.3s ~5.3s | |
- | Kandinsky 3.0 | 2.1B | INT4 Q8 FP16 | 61 tok/s 39 tok/s 22 tok/s | ~5.5s ~5.5s ~5.5s | |
- | LTX-2 | 2.2B | INT4 Q8 FP16 | 59 tok/s 38 tok/s 21 tok/s | ~5.7s ~5.7s ~5.8s | |
- | SDXL 1.0 Refiner | 2.31B | INT4 Q8 FP16 | 57 tok/s 36 tok/s 20 tok/s | ~6.0s ~6.0s ~6.0s | |
- | Stable Diffusion 3.5 Medium | 2.5B | INT4 Q8 FP16 | 54 tok/s 34 tok/s 19 tok/s | ~6.4s ~6.4s ~6.5s | |
- | SDXL 1.0 Base | 2.56B | INT4 Q8 FP16 | 53 tok/s 33 tok/s 18 tok/s | ~6.6s ~6.6s ~6.6s | |
- | SDXL Turbo | 2.56B | INT4 Q8 FP16 | 53 tok/s 33 tok/s 18 tok/s | ~6.6s ~6.6s ~6.6s | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 51 tok/s 30 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 49 tok/s 40 tok/s 26 tok/s | ~3.5s ~3.5s ~3.5s | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 49 tok/s 40 tok/s 26 tok/s | ~3.5s ~3.5s ~3.5s | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 49 tok/s 40 tok/s 26 tok/s | ~3.5s ~3.5s ~3.5s | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 48 tok/s 29 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
- | Mellum 2.1 12B A2.5B Thinking | 12B (2.5B active) | INT4 Q8 FP16 | 48 tok/s 32 tok/s 16 tok/s | ~7.4s ~7.4s ~8.7s | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 47 tok/s 29 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
- | LFM2-VL 3B | 3B | INT4 Q8 FP16 | 47 tok/s 29 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
- | Liquid D1 3B | 3B | INT4 Q8 FP16 | 47 tok/s 29 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
- | Liquid D1-VL 3B | 3B | INT4 Q8 FP16 | 47 tok/s 29 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
- | Liquid LFM2-VL 3B D1 | 3B | INT4 Q8 FP16 | 47 tok/s 29 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 47 tok/s 29 tok/s 16 tok/s | ~7.6s ~7.6s ~7.6s | |
- | LFM2.5 24B A2B | 24B (2.5B active) | INT4 | 43 tok/s | ~8.4s | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 | 40 tok/s 25 tok/s | ~9.3s ~10.1s | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 | 40 tok/s 25 tok/s | ~9.3s ~10.1s | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 | 38 tok/s | ~9.8s | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 | 38 tok/s | ~9.8s | |
- | FLUX.2 Klein 4B | 4B | INT4 Q8 FP16 | 37 tok/s 22 tok/s 12 tok/s | ~9.9s ~9.9s ~10.0s | |
- | FrogNano 4B | 4B | INT4 Q8 FP16 | 37 tok/s 22 tok/s 12 tok/s | ~9.9s ~9.9s ~10.0s | |
- | Hy-Memory 30B A3B | 30B (3B active) | INT4 | 37 tok/s | ~9.9s | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 | 37 tok/s | ~9.9s | |
- | Mage-VL Cowork | 4B | INT4 Q8 FP16 | 37 tok/s 22 tok/s 12 tok/s | ~9.9s ~9.9s ~10.0s | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 37 tok/s 22 tok/s 12 tok/s | ~9.9s ~9.9s ~10.0s | |
- | EmbeddingGemma 2 | 2B | INT4 Q8 FP16 | 35 tok/s 28 tok/s 18 tok/s | ~5.3s ~5.3s ~5.3s | |
- | MiniCPM-V 4.7 35B A3B | 35B (3B active) | INT4 | 35 tok/s | ~11.0s | |
- | CogVideoX 1.5 5B | 5B | INT4 Q8 FP16 | 31 tok/s 18 tok/s 10 tok/s | ~12.3s ~12.3s ~12.3s | |
- | CogVideoX 5B | 5B | INT4 Q8 FP16 | 31 tok/s 18 tok/s 10 tok/s | ~12.3s ~12.3s ~12.3s | |
- | Wan 2.2 TI2V 5B | 5B | INT4 Q8 FP16 | 31 tok/s 18 tok/s 10 tok/s | ~12.3s ~12.3s ~12.3s | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 30 tok/s 23 tok/s 14 tok/s | ~6.9s ~6.9s ~6.9s | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 29 tok/s 21 tok/s 13 tok/s | ~7.6s ~7.6s ~7.6s | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 29 tok/s 21 tok/s 13 tok/s | ~7.6s ~7.6s ~7.6s | |
- | AuraFlow v0.2 | 6.8B | INT4 Q8 FP16 | 24 tok/s 14 tok/s 7 tok/s | ~16.5s ~16.5s ~16.6s | |
- | AuraFlow v0.3 | 6.8B | INT4 Q8 FP16 | 24 tok/s 14 tok/s 7 tok/s | ~16.5s ~16.5s ~16.6s | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 24 tok/s 14 tok/s 7 tok/s | ~16.9s ~17.0s ~17.0s | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 23 tok/s 13 tok/s 7 tok/s | ~16.9s ~17.0s ~17.0s | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 23 tok/s 13 tok/s 7 tok/s | ~16.9s ~17.0s ~17.0s | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~17.6s ~17.7s ~17.7s | |
- | Youtu Parsing Omni | 8B | INT4 Q8 FP16 | 21 tok/s 12 tok/s 6 tok/s | ~19.3s ~19.3s ~19.4s | |
- | Stable Diffusion 3.5 Large | 8.1B | INT4 Q8 FP16 | 20 tok/s 12 tok/s 6 tok/s | ~19.5s ~19.5s ~19.6s | |
- | Stable Diffusion 3.5 Large Turbo | 8.1B | INT4 Q8 FP16 | 20 tok/s 12 tok/s 6 tok/s | ~19.5s ~19.5s ~19.6s | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 18 tok/s 12 tok/s 7 tok/s | ~14.6s ~14.6s ~14.7s | |
- | FLUX.2 Klein 9B | 9B | INT4 Q8 FP16 | 18 tok/s 11 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 18 tok/s 11 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 18 tok/s 11 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 18 tok/s 12 tok/s 7 tok/s | ~14.6s ~14.6s ~14.7s | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 17 tok/s 10 tok/s 5 tok/s | ~24.0s ~24.1s ~26.2s | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 | 17 tok/s 14 tok/s | ~8.9s ~9.7s | |
- | Mochi 1 Preview | 10B | INT4 Q8 FP16 | 17 tok/s 10 tok/s 5 tok/s | ~24.0s ~24.1s ~26.2s | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 16 tok/s 9 tok/s 4 tok/s | ~26.4s ~26.4s ~28.8s | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 16 tok/s 11 tok/s 6 tok/s | ~17.0s ~17.0s ~17.1s | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 16 tok/s 11 tok/s 6 tok/s | ~17.0s ~17.0s ~17.1s | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 16 tok/s 11 tok/s 6 tok/s | ~17.2s ~17.2s ~17.3s | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 16 tok/s 11 tok/s 6 tok/s | ~17.2s ~17.2s ~17.3s | |
- | FLUX.1 Canny Dev | 11.9B | INT4 Q8 FP16 | 14 tok/s 8 tok/s 4 tok/s | ~28.5s ~28.6s ~33.7s | |
- | FLUX.1 Depth Dev | 11.9B | INT4 Q8 FP16 | 14 tok/s 8 tok/s 4 tok/s | ~28.5s ~28.6s ~33.7s | |
- | FLUX.1 Dev | 11.9B | INT4 Q8 FP16 | 14 tok/s 8 tok/s 4 tok/s | ~28.5s ~28.6s ~33.7s | |
- | FLUX.1 Fill Dev | 11.9B | INT4 Q8 FP16 | 14 tok/s 8 tok/s 4 tok/s | ~28.5s ~28.6s ~33.7s | |
- | FLUX.1 Schnell | 11.9B | INT4 Q8 FP16 | 14 tok/s 8 tok/s 4 tok/s | ~28.5s ~28.6s ~33.7s | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 14 tok/s 10 tok/s 5 tok/s | ~19.3s ~19.3s ~19.4s | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 14 tok/s 10 tok/s 5 tok/s | ~19.3s ~19.3s ~19.4s | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 13 tok/s 9 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 13 tok/s 9 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 13 tok/s 9 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
- | HunyuanVideo 1.5 720P T2V | 13B | INT4 Q8 | 13 tok/s 7 tok/s | ~31.1s ~31.2s | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 13 tok/s 8 tok/s 5 tok/s | ~22.1s ~22.2s ~24.2s | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 13 tok/s 8 tok/s 5 tok/s | ~22.1s ~22.2s ~24.2s | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 13 tok/s 9 tok/s 5 tok/s | ~21.7s ~21.7s ~23.7s | |
- | Wan 2.1 I2V 14B 720P | 14B | INT4 Q8 | 12 tok/s 7 tok/s | ~33.5s ~33.5s | |
- | Wan 2.1 T2V 14B | 14B | INT4 Q8 | 12 tok/s 7 tok/s | ~33.5s ~33.5s | |
- | Wan 2.2 T2V A14B | 14B | INT4 Q8 | 12 tok/s 7 tok/s | ~33.5s ~33.5s | |
- | FLUX.2 Dev | 15B | INT4 Q8 | 11 tok/s 6 tok/s | ~35.8s ~35.8s | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 | 9 tok/s 6 tok/s | ~33.5s ~33.5s | |
- | Arete 27B | 27B | INT4 | 6 tok/s | ~63.6s | |
- | AREX-2 | 27B | INT4 | 6 tok/s | ~63.6s | |
- | PPLX Decider v1 27B | 27B | INT4 | 6 tok/s | ~63.6s | |
- | PPLX Decider v1.1 27B | 27B | INT4 | 6 tok/s | ~63.6s | |
- | Agnes 3.0 Flash | 33B | INT4 | 5 tok/s | ~77.4s | |
- | Falcon-40B | 40B | INT4 | 4 tok/s | ~101.6s | |
推理能力评估
不同量化精度下可运行的最大模型参数级别。
上下文长度:
4位量化
32B 模型
8位精度
14B 模型
16位全精度
8B 模型
微调能力评估
本地模型训练与适配器微调支持能力。
微调上下文:
QLoRA
32B 模型
LoRA
14B 模型
全参数微调
3B 模型
负载与选型建议
Apple M4 (32GB) 上的最佳量化精度与运行规格建议。
最佳推理区间
适合在全精度或 Q8 精度下运行 14B 模型，以及在 Q4 精度下搭配最高 32k 上下文运行 32B 模型。
显存带宽与推理速率
具备 120 GB/s 聚合带宽，单批次推理处于显存带宽受限模式。8B Q4 模型预估生成速度约 27 tok/s，70B Q4 模型预估生成速度约 3 tok/s。
统一内存配额
采用 75% 可用分配预算建模，以保留系统内存供 macOS 系统及图形渲染进程稳定运行。