选模型真实评测模型库编码 Agent 方法

公开数据导航

START HERE

PUBLIC DATA

按数据视角进入公开证据模型能力、Agent 与生成、成本与基础设施、数据与来源;每个数字都保留原始口径与出处。

MODEL INTELLIGENCE

模型能力比较模型、国家来源、公开评测与专业任务表现。

AGENT & GENERATION

Agent 与生成查看编码 Harness、Agent 能力与多媒体评测口径。

ECONOMICS & INFRA

成本与基础设施把能力放回调用成本、速度、部署条件中。

DATA & SOURCES

数据与来源公开数据目录、更新记录与方法口径。

模型对标

可从模型库、排行榜、评测页或模型详情页加入对标

数据快照：2026-09-19（基于公开数据）　当前页：2026-09-19

04

模型效果

在权威榜单与典型任务上的表现，结果均保留来源原始得分。

数据来源：公开评测与厂商官方数据完整度：17/25

只看差异

权威榜单（17 项）

Claude Opus 5 (max)

GPT-5.6 Sol (high)

Gemini 3.5 Flash历史版本

DeepSeek V4 Pro 0813 (max)

Artificial Analysis Intelligence Index分

50.7v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

42.5v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

33.0v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

36.3v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

GDPval-AA v2 归一化值%

66.1%v4.1 · 历史参考

55.8%v4.1 · 历史参考

42.0%v4.1 · 历史参考

53.9%v4.1 · 历史参考

GDPval-AA v2Elo

1,735.1裁判协议 v4.3.1Artificial Analysis · 2026-09-18T01:47:49.578Z

1,524.0裁判协议 v4.3.1Artificial Analysis · 2026-09-18T01:47:49.578Z

1,259.1裁判协议 v4.3.1Artificial Analysis · 2026-09-18T01:47:49.578Z

1,493.3裁判协议 v4.3.1Artificial Analysis · 2026-09-18T01:47:49.578Z

τ³-Banking%

42.1%v4.1 · 历史参考

36.7%v4.1 · 历史参考

32.2%v4.1 · 历史参考

39.6%v4.1 · 历史参考

Terminal-Bench v2.1%

89.1%v4.1 · 历史参考

87.3%v4.1 · 历史参考

78.7%v4.1 · 历史参考

78.7%v4.1 · 历史参考

Terminal-Bench v4.0%

49.0%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

20.7%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

6.6%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

14.1%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

AutomationBench-AA%

56.6%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

55.3%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

42.1%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

56.7%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

GDP.pdf%

21.6%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

27.8%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

19.8%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

11.4%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

AA-BriefcaseElo

1,633.4裁判协议 v4.3.1Artificial Analysis · 2026-09-18T01:47:49.578Z

1,360.9裁判协议 v4.3.1Artificial Analysis · 2026-09-18T01:47:49.578Z

869.7裁判协议 v4.3.1Artificial Analysis · 2026-09-18T01:47:49.578Z

1,264.7裁判协议 v4.3.1Artificial Analysis · 2026-09-18T01:47:49.578Z

SciCode%

56.4%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

57.8%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

53.9%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

51.0%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

AA-LCR%

79.3%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

81.7%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

73.3%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

80.3%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

AA-Omniscience分

37.1v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

20.4v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

21.2v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

0.8v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

AA-Omniscience 准确率%

60.9%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

58.4%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

51.4%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

49.1%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

AA-Omniscience 幻觉率%

60.8%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

91.2%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

62.2%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

94.8%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

Humanity's Last Exam%

54.9%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

46.0%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

42.7%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

41.0%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

GPQA Diamond%

93.2%v4.1 · 历史参考

92.8%v4.1 · 历史参考

92.2%v4.1 · 历史参考

92.8%v4.1 · 历史参考

CritPt%

29.1%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

25.7%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

13.1%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

18.0%v4.3Artificial Analysis · 2026-09-09T14:00:04.717Z

任务效果（8 项）

Claude Opus 5 (max)

GPT-5.6 Sol (high)

Gemini 3.5 Flash历史版本

DeepSeek V4 Pro 0813 (max)

Agent 能力指数分

暂无公开数据

48.5

37.4

36.4

编码能力指数分

暂无公开数据

77.2

70.1

59.4

金融与会计分

暂无公开数据

47.6

43.4

36.8

战略与运营分

暂无公开数据

47.7

41.4

38.1

法律分

暂无公开数据

48.9

49.3

36.9

医疗分

暂无公开数据

42.9

39.0

33.1

工程分

暂无公开数据

56.8

51.5

45.8

经济学分

暂无公开数据

52.5

52.3

42.7