模型实测一览

本站按模型与 Mac 整理的实测,独立于用途推荐。量化、输入和运行环境不同,结果也会改变。

前往用途推荐 →

文本模型

每台 Mac、每个模型采用最新保存记录。回复时间为简短 JCommonsenseQA 的中位数,生成速度来自另一生成任务;二者任务不同。箭头表示有利方向,不是综合排名。

如何阅读此表

使用内存是运行环境报告的模型用量(估计值单独标注),不是整机峰值或所需 RAM。交换增量是新增 SSD 交换量,不代表流畅度。正确率附答对/总题数,请仅比较相容条件。

28 条记录

可横向滚动,模型名固定在左侧。

文本模型
ModelMac测量日期 (JST)回复时间
秒 ↓
生成速度
tok/s ↑
使用内存
GB ↓
交换增量
MB ↓
知识
JMMLU ↑
算术
MGSM ↑
条件与详情
gemma3:27bQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-020.3823.1817.520.276.0%152 / 20091.0%91 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_k": 64,
      "top_p": 0.95
    }
  },
  "digest": "6be2e2ba6843b9255820e2de2cbbe75cfdef0294ac7a531572e654d6e0eefe7d"
}
条件与详情 → (English)
gemma3:27bQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-170.2546.9816.76074.5%149 / 20093.0%93 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_k": 64,
      "top_p": 0.95
    }
  },
  "digest": "6be2e2ba6843b9255820e2de2cbbe75cfdef0294ac7a531572e654d6e0eefe7d"
}
条件与详情 → (English)
gemma4:12bQ4_K_MApple M6 · 16 GBApple M6 · 16 GB2026-10-2426.3216.258.13226.1未测量未测量
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.0",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_k": 64,
      "top_p": 0.95
    }
  },
  "digest": "4eb23ef187e2c5462566d6a1d3bbbc2f1346d0b4327cbb66d58fffbcc9b2b05c"
}
条件与详情 → (English)
gemma4:26bQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-023.18103.3918.73估计值085.0%170 / 20098.0%98 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_k": 64,
      "top_p": 0.95
    }
  },
  "digest": "001e5dafc3c77684c2307ebc6ab8e336e10c9b18eca52acf547d72fc83c3ca8c"
}
条件与详情 → (English)
gemma4:26bQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-182.1159.4618.73估计值085.5%171 / 20097.0%97 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "top_k": 64,
      "top_p": 0.95,
      "temperature": 1.0
    }
  },
  "digest": "001e5dafc3c77684c2307ebc6ab8e336e10c9b18eca52acf547d72fc83c3ca8c"
}
条件与详情 → (English)
glm-4.7-flash:latestQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-027.5481.0119.42076.0%152 / 20095.0%95 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "top_p": 0.95,
      "min_p": 0.01,
      "repeat_penalty": 1.0,
      "temperature": 1.0
    }
  },
  "digest": "4475827791a269b02c8ec49b1c3bc1abb5846bacf3fae015b75d33986322d8f6"
}
条件与详情 → (English)
glm-4.7-flash:latestQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-175.79101.8319.54082.5%165 / 20095.0%95 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_p": 0.95,
      "min_p": 0.01,
      "repeat_penalty": 1.0
    }
  },
  "digest": "4475827791a269b02c8ec49b1c3bc1abb5846bacf3fae015b75d33986322d8f6"
}
条件与详情 → (English)
gpt-oss:120bMXFP4Apple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-181.88114.7964.33082.0%164 / 20099.0%99 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0
    }
  },
  "digest": "714d86958b6e870a9320e78126ac1ce3477727a1cd9bf7194b8ace0d84c62fd8"
}
条件与详情 → (English)
gpt-oss:20bMXFP4Apple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-021.7789.8712.84083.0%166 / 20098.0%98 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0
    }
  },
  "digest": "f38aa0c53da5f8c49d08c43a99df24ff53167fe68e24664a7777288e7656fdfe"
}
条件与详情 → (English)
gpt-oss:20bMXFP4Apple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-171.04164.7212.86081.0%162 / 20097.0%97 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0
    }
  },
  "digest": "f38aa0c53da5f8c49d08c43a99df24ff53167fe68e24664a7777288e7656fdfe"
}
条件与详情 → (English)
hf.co/openbmb/MiniCPM5-2B-GGUF:Q4_K_MQ4_K_MApple M6 · 16 GBApple M6 · 16 GB2026-10-122.8881.552.05070.0%140 / 20090.0%90 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.0",
  "sampling": {
    "source": "generation_config.json(openbmb/MiniCPM5-2B)",
    "options": {
      "temperature": 1.0,
      "top_p": 0.95
    }
  },
  "digest": "e680c6c9f9033490cb68df7c1145f733a3a1b377b387af0937251a70d3c71a14"
}
条件与详情 → (English)
MiniCPM5-2BQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-021.5187.642.02073.0%146 / 20090.0%90 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "generation_config.json(openbmb/MiniCPM5-2B)",
    "options": {
      "temperature": 1.0,
      "top_p": 0.95
    }
  },
  "digest": "e680c6c9f9033490cb68df7c1145f733a3a1b377b387af0937251a70d3c71a14"
}
条件与详情 → (English)
hf.co/openbmb/MiniCPM5-2B-GGUF:Q4_K_MQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-181.07247.241.66071.5%143 / 20089.0%89 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "generation_config.json(openbmb/MiniCPM5-2B)",
    "options": {
      "temperature": 1.0,
      "top_p": 0.95
    }
  },
  "digest": "e680c6c9f9033490cb68df7c1145f733a3a1b377b387af0937251a70d3c71a14"
}
条件与详情 → (English)
Hy-Memory-30B-A3BQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-020.1110.3919.22065.5%131 / 20088.0%88 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "なし(温度0)",
    "options": {
      "temperature": 0
    }
  },
  "digest": "3d1cd167746ae18e21cbbb8f5200d524caa0d5239470e94b8caafc2931a15c46"
}
条件与详情 → (English)
hf.co/tencent/Hy-Memory-30B-A3B-GGUF:Q4_K_MQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-220.08142.0118.89068.5%137 / 20090.0%90 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "なし(温度0)",
    "options": {
      "temperature": 0
    }
  },
  "digest": "3d1cd167746ae18e21cbbb8f5200d524caa0d5239470e94b8caafc2931a15c46"
}
条件与详情 → (English)
hf.co/unsloth/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF:Q4_K_MQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-195.64140.8723.88081.5%163 / 20098.0%98 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "なし(温度0)",
    "options": {
      "temperature": 0
    }
  },
  "digest": "ad2ef3b478c378987150cc477ce4ca3492f9c91572f36fa3c4f162c4796a9cf2"
}
条件与详情 → (English)
muse-glimmer:30bQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-0215.0124.8517.58-283.183.0%166 / 20098.0%98 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_k": 64,
      "top_p": 0.95
    }
  },
  "digest": "de878ce33ad81d060001db1469a02eebe4d86f0ad58cfe52dc062fdcbe4464c1"
}
条件与详情 → (English)
muse-glimmer:30bQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-187.5251.2215.8084.0%168 / 20098.0%98 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_k": 64,
      "top_p": 0.95
    }
  },
  "digest": "de878ce33ad81d060001db1469a02eebe4d86f0ad58cfe52dc062fdcbe4464c1"
}
条件与详情 → (English)
qwen3-coder-next:latestQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-180.12107.1952.05076.5%153 / 20095.0%95 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_k": 40,
      "top_p": 0.95
    }
  },
  "digest": "ca06e9e4087c714d44355bf954099187890e63084b4a632b8e9956c4b9492074"
}
条件与详情 → (English)
qwen3-coder:30bQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-020.09107.719.36069.0%138 / 20089.0%89 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 0.7,
      "top_k": 20,
      "top_p": 0.8,
      "repeat_penalty": 1.05
    }
  },
  "digest": "06c1097efce0431c2045fe7b2e5108366e43bee1b4603a7aded8f21689e90bca"
}
条件与详情 → (English)
qwen3-coder:30bQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-170.07160.4819.37070.5%141 / 20094.0%94 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "repeat_penalty": 1.05,
      "temperature": 0.7,
      "top_k": 20,
      "top_p": 0.8
    }
  },
  "digest": "06c1097efce0431c2045fe7b2e5108366e43bee1b4603a7aded8f21689e90bca"
}
条件与详情 → (English)
qwen3-next:80bQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-187.996.650.66085.5%171 / 20098.0%98 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "top_k": 20,
      "top_p": 0.95,
      "repeat_penalty": 1.0,
      "temperature": 0.6
    }
  },
  "digest": "b2ebb986e4e960bb3f2d636472a466b4027939c4d8d9a62fc82bed8ebfd1ae19"
}
条件与详情 → (English)
qwen3.5:4bQ4_K_MApple M6 · 16 GBApple M6 · 16 GB2026-10-3063.5644.153.360未测量未测量
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.0",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "presence_penalty": 1.5,
      "temperature": 1.0,
      "top_k": 20,
      "top_p": 0.95
    }
  },
  "digest": "3c3adf637187c95955232acd3ed05b9ea45aeb2a5f953e631c38174a46b5b56a"
}
条件与详情 → (English)
qwen3.5:9bnvfp4Apple M6 · 16 GBApple M6 · 16 GB2026-11-0147.8433.447.68946未测量未测量
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.40.13",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "min_p": 0.0,
      "presence_penalty": 1.5,
      "repeat_penalty": 1.0,
      "temperature": 1.0,
      "top_k": 20,
      "top_p": 0.95
    }
  },
  "digest": "d63f830f28d515fefea546f213fca7a75e9dd5176ef51a6deabc952d1c99a911"
}
条件与详情 → (English)
qwen3.6:35b-a3bQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-028.52125.7122.272949.286.5%173 / 20098.0%98 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "top_k": 20,
      "top_p": 0.95,
      "min_p": 0.0,
      "presence_penalty": 1.5,
      "repeat_penalty": 1.0,
      "temperature": 1.0
    }
  },
  "digest": "a7eb95c53bcf96b4bdd008d0fab4a5dac88047d9c1a7a9ab88ed453423fbd87c"
}
条件与详情 → (English)
qwen3.6:35b-a3bQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-188.04137.8721.7087.5%175 / 200100.0%100 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "temperature": 1.0,
      "top_k": 20,
      "top_p": 0.95,
      "min_p": 0.0,
      "presence_penalty": 1.5,
      "repeat_penalty": 1.0
    }
  },
  "digest": "a7eb95c53bcf96b4bdd008d0fab4a5dac88047d9c1a7a9ab88ed453423fbd87c"
}
条件与详情 → (English)
qwen3.8:27bQ4_K_MApple M5 Max · 36 GBApple M5 Max · 36 GB2026-11-027.6926.2918.151208.586.5%173 / 200100.0%100 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.2",
  "Ollama": "0.34.3",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "top_k": 20,
      "top_p": 0.95,
      "min_p": 0.0,
      "presence_penalty": 0.0,
      "repeat_penalty": 1.0,
      "temperature": 1.0
    }
  },
  "digest": "aaee06c39dcf2437cde036998d960e1fc1494b8191be7cc9657d01e509097813"
}
条件与详情 → (English)
qwen3.8:27bQ4_K_MApple M5 Ultra · 96 GBApple M5 Ultra · 96 GB2026-10-183.7955.1416.43085.5%171 / 200100.0%100 / 100
条件与详情
{
  "num_ctx": 8192,
  "benchmark_ctx": 20480,
  "runs": 3,
  "temperature": 0,
  "seed": 42,
  "method_version": 1,
  "datasets": {
    "mgsm_ja": "MGSM-Rev2"
  },
  "benchmark_limits": {
    "jmmlu": 16384,
    "jcommonsenseqa": 4096,
    "mgsm_ja": 16384
  },
  "macOS": "27.0.1",
  "Ollama": "0.40.5",
  "sampling": {
    "source": "Ollama 公式タグ",
    "options": {
      "min_p": 0.0,
      "presence_penalty": 0.0,
      "repeat_penalty": 1.0,
      "temperature": 1.0,
      "top_k": 20,
      "top_p": 0.95
    }
  },
  "digest": "aaee06c39dcf2437cde036998d960e1fc1494b8191be7cc9657d01e509097813"
}
条件与详情 → (English)

测试方法

实用测试:模型与任务记录

已公开的代表性记录,不是所有模型的同条件对比。合格与否针对各任务标准,不代表模型整体质量,详情保留条件与历史记录。

可横向滚动,模型名固定在左侧。

实用测试:模型与任务记录
Model任务Mac测量日期 (JST)生成时间(秒)结果条件与详情
Qwen3.8 27B (Q4_K_M)修复一处链接Mac Studio · M5 Ultra · 96GB2026-11-175.34合格 · 21/21条件与详情 →
mlx-community/whisper-large-v3-turbo日语语音识别Mac mini2026-11-0618.17不合格 · CER 10.5%条件与详情 →
mlx-community/whisper-large-v3-mlx日语语音识别Mac mini2026-11-088.68不合格 · CER 26.3%条件与详情 →
qwen3.6:35b-a3b整理现有转录中的数字(不含语音识别)Mac Studio · M5 Ultra · 96GB2026-11-158.36合格条件与详情 →
qwen3.8:27b从会议记录提取五项答案与证据Mac Studio · M5 Ultra · 96GB2026-11-1713.00合格条件与详情 →

图像与视频模型

图像和视频与文本任务不同。图像时间为后续图片,视频时间为单段;内存是各后端报告的峰值,不可跨实现直接比较。

可横向滚动,模型名固定在左侧。

图像与视频模型
Model图像与视频模型Mac测量日期 (JST)生成时间(秒)峰值内存(GB)条件与详情
Z-Image Turbo图像Apple M5 Max · 36 GB2026-09-2419.718.51
条件与详情

8bit · v0.20.0

{
  "width": 1024,
  "height": 1024,
  "seeds": [
    1,
    2
  ],
  "mlx_cache_limit_gb": 2
}
条件与详情 → (English)
FLUX.2 [klein] 4B图像Apple M5 Max · 36 GB2026-09-24516.67
条件与详情

8bit · v0.20.0

{
  "width": 1024,
  "height": 1024,
  "seeds": [
    1,
    2
  ],
  "mlx_cache_limit_gb": 2
}
条件与详情 → (English)
ERNIE-Image Turbo图像Apple M5 Max · 36 GB2026-09-2422.821.63
条件与详情

8bit · v0.20.0

{
  "width": 1024,
  "height": 1024,
  "seeds": [
    1,
    2
  ],
  "mlx_cache_limit_gb": 2
}
条件与详情 → (English)
Qwen-Image图像Apple M5 Max · 36 GB2026-09-2412737.3
条件与详情

4bit · v0.20.0

{
  "width": 1024,
  "height": 1024,
  "seeds": [
    1,
    2
  ],
  "mlx_cache_limit_gb": 2
}
条件与详情 → (English)
Wan 2.2 TI2V-5B视频Apple M5 Max · 36 GB2026-09-24852.745.72
条件与详情

— ·

{
  "width": 1280,
  "height": 704,
  "frames": 81,
  "fps": 24,
  "seed": 1,
  "mlx_cache_limit_gb": 2,
  "seconds": 3.4
}
条件与详情 → (English)