科研

偏科学知识、数学推理与研究型问题求解。更新于 07/16 02:55

模型

160 个 · 16 项基准

评分

HLE (Humanity's Last Exam)

Arena Text Rating

FrontierMath Tier 4

FrontierMath Tier 1-3

Frontier Science Research

HLE-Full (w/ Tools)

HealthBench Professional

100.0

93.6%

46.9%

52.2%

1482.3 Elo

72.5%

85.3%

1769 Elo

92.1%

98.3%

33.9%

96.5%

44.8%

52.2%

96.7%

51.8%

27.1%

2 Claude Opus 4.8

99.1

93.6%

49.8%

57.9%

1483.9 Elo

56.1%

80%

1890 Elo

83.5%

95.7%

32.9%

96.5%

49.8%

57.9%

96.7%

56.9%

20.9%

3 Claude Fable 5

98.5

92.6%

-

-

1508.5 Elo

87.8%

87%

1932 Elo

-

-

-

-

59%

64.5%

-

66%

-

97.8

-

43.1%

57.2%

-

39.6%

52.4%

-

-

-

-

-

-

-

-

-

-

5 Gemini 3.1 Pro

97.2

94.3%

45%

51.6%

1485.8 Elo

16.7%

59.6%

1314 Elo

91%

98.3%

16.7%

94.8%

45.4%

51.4%

94.7%

41.6%

17.7%

96.9

92.4%

41.4%

53.5%

-

-

-

-

90%

97%

24.8%

95%

-

-

97.1%

-

13.4%

96.9

94.4%

42.7%

58.7%

-

38%

50%

-

-

-

36.7%

-

-

-

-

-

-

96.7

94.6%

-

-

1484.3 Elo

65.9%

86%

-

-

-

-

-

-

-

-

60.5%

-

95.4

93%

39.8%

52.1%

1477.8 Elo

27.1%

47.6%

1674 Elo

91.4%

99.2%

33%

95.8%

39.8%

52.1%

97.7%

-

-

10 Muse Spark 1.1

95.2

-

-

-

1491.0 Elo

-

-

1381 Elo

-

-

-

-

52.2%

62.1%

-

59.3%

-

11 Claude Opus 4.6

94.2

91.3%

40%

53.1%

1503.6 Elo

-

-

1619 Elo

75.3%

96.7%

-

96.3%

40%

53%

96.2%

-

12.6%

93.8

91.2%

40.5%

54.7%

1468.7 Elo

-

-

-

91%

99.2%

19.8%

94.4%

-

-

92.5%

-

20.9%

13 DeepSeek-V4-Pro

93.3

90.1%

37.7%

48.2%

1457.1 Elo

-

-

1554 Elo

93%

94.6%

21.3%

94.4%

-

-

95.2%

-

14%

14 Gemini 3 Pro

93.1

91.9%

37.5%

45.8%

1485.8 Elo

-

-

-

83.3%

90.6%

-

93.3%

-

-

-

-

-

93.1

91%

36.4%

54%

1461.7 Elo

-

-

1482 Elo

91.1%

96.4%

-

-

34.7%

54%

92.7%

-

9.1%

16 Claude Opus 4.7

92.5

94.2%

46.9%

54.7%

1502.9 Elo

22.9%

43.8%

-

81.8%

-

-

-

-

-

-

-

-

92.3

93.2%

36.6%

50%

-

31.3%

-

-

-

-

-

-

-

-

-

-

-

18 GPT-5.6 Terra

92.3

92.9%

-

-

-

68.3%

84.9%

-

-

-

-

-

-

-

-

57.7%

-

92.1

90.4%

37%

53.2%

-

-

-

-

90%

-

21.3%

-

-

-

-

-

-

20 Qwen3.7-Plus

91.7

90.3%

34.7%

-

1462.5 Elo

-

-

-

86%

-

-

-

-

-

92.9%

-

6%

91.0

92.4%

35.5%

45.5%

1437.5 Elo

18.8%

40.7%

-

86.3%

96.7%

25.2%

100%

-

-

-

-

-

22 Claude Sonnet 5

91.0

-

-

-

1464.3 Elo

-

-

1618 Elo

-

-

-

-

43.2%

57.4%

-

-

-

90.7

89.8%

35.9%

55.7%

-

-

-

-

88.5%

-

28.3%

-

-

-

-

-

-

90.3

-

-

-

1487.6 Elo

-

-

1145 Elo

-

-

-

-

42.8%

50.4%

-

54.1%

-

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