本地

可本地部署、参数量小于 40B 的前沿模型集合。更新于 07/16 02:55

模型

47 个 · 16 项基准

评分

LiveCodeBench-v6

HLE (Humanity's Last Exam)

Arena Text Rating

LiveBench 20241125

Arena WebDev Rating

OmniDocBench 1.5

100.0

87.8%

93.5%

86.2%

83.9%

66%

24%

-

94.1%

90.7%

93.8%

-

-

-

-

-

-

93.8

85.5%

93.2%

86.1%

80.7%

65.6%

24.3%

1408.8 Elo

-

89.8%

92%

-

-

-

1356.8 Elo

-

-

3 Qwen3.6-35B-A3B

93.2

86%

93.3%

85.2%

80.4%

64.7%

21.4%

-

92.7%

89.1%

90.7%

-

-

-

-

75.3%

89.9%

4 Qwen3-30B-A3B-Thinking-2507

89.7

73.4%

91.4%

80.9%

66%

56.8%

-

-

-

-

-

85%

72.4%

76.8%

-

-

-

89.7

84.3%

-

85.2%

80%

-

19.5%

1451.1 Elo

89.2%

-

-

-

-

-

1371.4 Elo

76.9%

0.1%

6 Qwen3.5-35B-A3B

87.6

84.2%

93.3%

85.3%

74.6%

63.4%

22.4%

1395.6 Elo

91%

89.2%

89%

-

-

-

1249.6 Elo

75.1%

89.3%

86.1

68.4%

90.9%

-

-

-

-

1347.3 Elo

-

-

-

72.9%

70.3%

74.9%

-

-

-

85.5

81.7%

91.1%

82.5%

-

58.2%

-

-

-

-

-

-

-

-

-

-

-

9 Gemma 4 26B A4B

85.2

82.3%

-

82.6%

77.1%

-

8.7%

1438.2 Elo

88.3%

-

-

-

-

-

1360.7 Elo

73.8%

0.1%

84.2

64%

88.6%

-

-

-

-

-

-

-

-

70.4%

70.4%

71.3%

-

-

-

11 Qwen3-4B-Thinking-2507

82.9

65.8%

86.1%

74%

55.2%

47.8%

-

-

-

-

-

81.3%

71.2%

71.8%

-

-

-

12 Qwen3-30B-A3B

82.8

73.4%

91.4%

80.9%

29%

56.8%

-

1327.3 Elo

-

-

-

70.9%

69.1%

74.3%

-

-

-

82.6

65.6%

90%

-

-

-

-

1336.2 Elo

-

-

-

69.5%

66.4%

72%

-

-

-

14 Qwen3-30B-A3B-Instruct-2507

82.2

70.4%

89.3%

78.4%

43.2%

53.4%

-

1383.2 Elo

-

-

-

61.3%

65.1%

69%

-

-

-

81.6

76.2%

88.8%

79.1%

-

52.9%

-

-

-

-

-

-

-

-

-

-

-

78.8

62%

87.5%

-

-

-

-

-

-

-

-

67.3%

68.1%

67.1%

-

-

-

78.5

71.5%

87.8%

74.8%

-

48.5%

-

1317.4 Elo

-

-

-

-

-

-

-

-

-

18 DeepSeek-R1-Distill-Qwen-32B

75.8

62.1%

88.2%

-

-

-

-

-

-

-

-

49.6%

53.5%

45.6%

-

-

-

19 Qwen3-4B-Instruct-2507

75.4

62%

84.2%

69.6%

35.1%

42.8%

-

-

-

-

-

47.4%

61.9%

63%

-

-

-

73.3

58.6%

-

69.4%

52%

-

-

-

42.5%

-

-

-

-

-

-

52.6%

0.2%

21 DeepSeek-R1-Distill-Qwen-14B

73.0

59.1%

84.1%

-

-

-

-

-

-

-

-

44.5%

49.5%

52.3%

-

-

-

72.9

55.9%

83.7%

58%

26.4%

32%

-

-

-

-

-

65.6%

65.9%

63.6%

-

-

-

23 DeepSeek-R1-Distill-Qwen-7B

71.3

49.1%

-

-

-

-

-

-

-

-

-

-

-

-

-

-

-

71.3

-

-

-

-

-

-

1303.3 Elo

-

-

-

-

-

-

-

-

-

低高

单一有效来源基准按 0.5 权重计入，Arena.ai 除外；最高模型归一化为 100