开源

仅聚合已标记为开源的模型，综合推理、编程、Agent、多模态与长上下文表现。更新于 07/16 02:55

模型

108 个 · 16 项基准

评分

HLE (Humanity's Last Exam)

SWE-bench Verified

SWE-bench Multilingual

Arena WebDev Rating

Arena Text Rating

Terminal Bench 2.1 (Terminus-2)

Claw Eval (pass^3)

Arena Agent Bash Recovery

100.0

76.8%

62.1%

91.2%

40.5%

84.2%

48.9%

83%

91%

1584.3 Elo

1468.7 Elo

-

81%

73.4%

54.7%

62.4%

5.2%

95.6

79.1%

57.9%

90.4%

37%

78%

45.6%

75.8%

90%

-

-

84.2%

71.7%

73.4%

53.2%

68.5%

-

3 Kimi-K2.7-Code

95.5

76%

-

-

-

-

-

-

-

1470.6 Elo

-

-

-

-

-

-

-0.2%

4 DeepSeek-V4-Pro

93.7

73.6%

59%

90.1%

37.7%

80.6%

41.5%

76.5%

93%

1458.2 Elo

1457.1 Elo

83.4%

64%

71.3%

48.2%

58.4%

4.5%

93.5

69.4%

59.5%

91%

36.4%

80.2%

42.8%

77.1%

91.1%

1513.8 Elo

1461.7 Elo

83.2%

67.2%

70.2%

54%

62.3%

-1.8%

90.1

74.2%

59%

93%

37%

80.5%

42.1%

-

-

1501.3 Elo

1446.5 Elo

83.5%

66%

-

-

74.5%

6.3%

89.9

71.8%

58.8%

86.2%

34.7%

76.2%

42.7%

74.8%

86.8%

1527.4 Elo

1472.2 Elo

79.3%

63.5%

71%

52.3%

62.7%

4.3%

8 DeepSeek-V4-Flash

86.8

69%

52.6%

88.5%

34.8%

79%

39.4%

75%

91.1%

-

1438.2 Elo

73.2%

54.2%

64.8%

45.1%

61.8%

4.1%

9 Qwen3.5-397B-A17B

86.7

74.2%

50.9%

88.4%

28.7%

76.4%

32.2%

70.9%

83.1%

1396.9 Elo

1443.0 Elo

69%

49.9%

68.7%

48.3%

-

-

85.9

-

59.5%

88.9%

-

-

-

77.3%

81.8%

-

-

79.9%

-

-

-

-

-

85.7

69.8%

55.1%

86%

30.5%

77.8%

35.9%

73.3%

82.5%

1430.2 Elo

1457.0 Elo

62%

-

63.3%

50.4%

-

-

12 MiMo-V2.5-Pro

85.3

-

-

-

-

-

-

-

-

1475.0 Elo

1466.2 Elo

-

-

-

-

-

3.6%

84.1

-

56.1%

-

-

-

-

-

-

1429.7 Elo

1433.2 Elo

-

-

-

-

62.3%

-

83.7

63.8%

53.8%

87.6%

31.5%

76.8%

32%

73%

81.8%

1433.4 Elo

1449.7 Elo

74.9%

-

70%

51.8%

52.3%

-

83.2

-

53.5%

87.8%

24%

77.2%

36.2%

71.3%

80.8%

-

-

-

-

-

-

-

-

16 Nemotron-3-Ultra

82.3

-

-

87%

26.7%

70.7%

-

67.7%

88.6%

-

-

44.4%

56.4%

65.4%

37.4%

-

-15.3%

17 Qwen3.5-122B-A10B

81.6

-

-

86.6%

25.3%

-

-

-

-

1363.9 Elo

1417.1 Elo

-

-

66.9%

-

-

-

80.5

52%

-

85.7%

24.8%

73.8%

-

66.7%

82%

1440.3 Elo

1442.4 Elo

52%

-

-

42.8%

-

-

79.3

-

-

85.5%

24.3%

-

-

-

-

1356.8 Elo

1408.8 Elo

-

-

66.1%

-

-

-

20 DeepSeek-V3.2

78.6

62.2%

-

82.4%

25.1%

73.1%

-

70.2%

78.3%

1367.9 Elo

1425.1 Elo

51.4%

-

-

40.8%

-

-

21 Qwen3-235B-A22B-Thinking-2507

78.6

-

-

81.1%

18.2%

-

-

-

-

-

1399.2 Elo

-

-

-

-

-

-

22 Qwen3.6-35B-A3B

78.1

62.8%

49.5%

86%

21.4%

73.4%

29.4%

67.2%

78.9%

-

-

-

-

-

-

-

-

78.0

-

-

84.3%

19.5%

-

-

-

-

1371.4 Elo

1451.1 Elo

-

-

-

-

-

-28.7%

24 MiniMax M2.7

77.2

49.4%

56.2%

87%

28%

79.9%

39.8%

71.8%

68.3%

1396.1 Elo

1416.8 Elo

76.3%

55.5%

69.8%

-

49.7%

-4.7%

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