总体

综合用户偏好、通用知识与指令遵循能力。更新于 07/16 02:55

模型

164 个 · 16 项基准

评分

Arena Agent Bash Recovery

Arena Agent Confirmed Success

Arena Agent Net Improvement

Arena Agent Praise vs Complaint

Arena Agent Steerability

Arena Agent Tool Hallucination

FrontierMath Tier 4

FrontierMath Tier 1-3

Arena Text Rating

Arena WebDev Rating

Terminal-Bench 2.1

1 Claude Fable 5

100.0

80.4%

88%

92.6%

9.4%

17.3%

13.9%

30.6%

12.1%

1.3%

87.8%

87%

1508.5 Elo

1652.7 Elo

70%

84.3%

61.7%

99.6

64.6%

90.4%

94.6%

7.5%

10.9%

10.9%

17.6%

17.3%

1.3%

65.9%

86%

1484.3 Elo

1631.0 Elo

72.7%

88.8%

58%

3 Claude Opus 4.8

94.9

69.2%

84.3%

93.6%

9.9%

8.6%

9.4%

18.0%

10.3%

0.6%

56.1%

80%

1483.9 Elo

1560.7 Elo

59%

82.7%

59.9%

94.4

59.4%

84.4%

93.6%

14.5%

6.9%

8.3%

13.6%

6.7%

1.3%

72.5%

85.3%

1482.3 Elo

1501.8 Elo

67%

85.6%

55.6%

92.8

-

90.1%

-

-

-

-

-

-

-

39.6%

52.4%

-

-

-

-

-

6 GPT-5.6 Terra

92.3

63.4%

87.5%

92.9%

-

-

-

-

-

-

68.3%

84.9%

-

-

69.6%

87.4%

53.1%

90.7

64.7%

-

-

9.7%

6.9%

5.0%

6.1%

0.9%

1.3%

-

-

1463.9 Elo

1557.0 Elo

53%

83.3%

-

90.4

-

89.3%

94.4%

-

-

-

-

-

-

38%

50%

-

-

-

-

-

90.3

62.1%

-

91.2%

5.2%

9.1%

6.9%

15.4%

4.1%

1.3%

-

-

1468.7 Elo

1584.3 Elo

44%

-

48.2%

10 Claude Sonnet 5

90.2

63.2%

-

-

9.6%

10.3%

8.0%

13.9%

4.9%

1.2%

-

-

1464.3 Elo

1551.3 Elo

-

-

-

11 Claude Opus 4.7

89.7

64.3%

79.3%

94.2%

11.7%

5.5%

8.2%

13.7%

9.2%

1.3%

22.9%

43.8%

1502.9 Elo

1558.9 Elo

-

78.9%

-

88.8

57.7%

82.7%

93%

11.3%

6.6%

6.7%

6.1%

8.0%

1.3%

27.1%

47.6%

1477.8 Elo

1456.9 Elo

-

-

54.6%

13 Muse Spark 1.1

88.4

61.5%

-

-

4.0%

2.1%

0.2%

0.6%

-7.2%

1.3%

-

-

1491.0 Elo

1535.8 Elo

53.3%

80%

-

87.2

60.6%

-

92.4%

6.8%

-2.3%

-0.8%

-5.4%

-4.2%

0.9%

-

-

-

-

-

-

45.1%

15 Claude Opus 4.6

86.8

57.3%

84%

91.3%

11.8%

3.5%

6.5%

9.4%

6.7%

1.3%

-

-

1503.6 Elo

1542.1 Elo

-

-

47.2%

86.7

57.9%

84.2%

90.4%

-

-

-

-

-

-

-

-

-

-

-

-

48.5%

17 GPT-5.6 Luna

86.5

62.7%

83.3%

92.3%

-

-

-

-

-

-

58.5%

78.6%

-

-

67.2%

84.7%

53.4%

86.4

57.5%

86.2%

89.8%

-

-

-

-

-

-

-

-

-

1538.9 Elo

-

71%

50.6%

19 Kimi-K2.7-Code

86.1

-

-

-

-0.2%

4.5%

-0.8%

1.4%

-1.6%

1.3%

-

-

-

1470.6 Elo

-

-

-

20 Gemini 3.1 Pro

86.0

54.2%

85.9%

94.3%

-6.3%

0.6%

-0.7%

-0.1%

1.8%

1.3%

16.7%

59.6%

1485.8 Elo

1445.4 Elo

12%

70.7%

48.8%

21 Qwen3.7-Plus

85.2

57.6%

-

90.3%

4.6%

-1.1%

-0.6%

-4.0%

-2.7%

0.2%

-

-

1462.5 Elo

-

-

-

-

85.0

59.5%

83.2%

91%

-1.8%

2.4%

-1.8%

-0.2%

-4.1%

1.3%

-

-

1461.7 Elo

1513.8 Elo

-

-

50%

23 DeepSeek-V4-Pro

84.3

59%

83.4%

90.1%

4.5%

-3.0%

-1.1%

-4.5%

-3.1%

0.8%

-

-

1457.1 Elo

1458.2 Elo

-

-

51.8%

24 Claude Sonnet 4.6

83.9

58.1%

74.7%

-

11.6%

-0.9%

2.2%

-0.8%

1.3%

1.3%

-

-

1472.2 Elo

1521.5 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