Claude Sonnet 4 Benchmark Details

Claude Sonnet 4 currently shows benchmark results led by SWE-bench Verified (14 / 115, score 80.20), Terminal-Bench (10 / 35, score 41.30), MMLU Pro (39 / 133, score 84). 1 source link is attached for reference.

Benchmark Results

Claude Sonnet 4

Benchmark Results

General Knowledge

9 evaluations

Benchmark / mode

Score

Rank/total

MMLU Pro

Thinking Enabled

39 / 133

LiveBench

Standard Mode

50.98

89 / 115

LiveBench

64K

61.27

65 / 115

ARC-AGI-1

Standard Mode

23.80

80 / 92

ARC-AGI-1

Thinking Enabled

73 / 92

HLE

Standard Mode

5.52

180 / 189

HLE

Thinking Enabled

9.60

166 / 189

ARC-AGI-2

Standard Mode

1.30

78 / 85

ARC-AGI-2

Thinking Enabled

5.90

69 / 85

Other

3 evaluations

Benchmark / mode

Score

Rank/total

GPQA Diamond

Standard Mode

204 / 270

GPQA Diamond

Thinking Enabled

75.40

167 / 270

GPQA Diamond

DeepTools

83.80

111 / 270

Coding and Software Engineer

6 evaluations

Benchmark / mode

Score

Rank/total

CodeClash

Standard ModeTools

1223

4 / 8

SWE-bench Verified

Thinking EnabledTools

72.70

52 / 115

SWE-bench Verified

Thinking Level · HighTools

80.20

14 / 115

LiveCodeBench

Standard Mode

48.50

100 / 127

LiveCodeBench

Thinking Enabled

62 / 127

SWE-Bench Pro - Public

Thinking Enabled

42.70

54 / 60

Math and Reasoning

12 evaluations

Benchmark / mode

Score

Rank/total

AIME2025

Standard Mode

95 / 106

AIME2025

Thinking Enabled

70.50

71 / 106

AIME2025

DeepTools

50 / 106

AIME 2024

Standard Mode

43.40

50 / 62

IMO-ProofBench

Thinking Enabled

27.10

8 / 16

IMO 2024

Standard Mode

9.70

5 / 10

IMO 2024

Thinking Enabled

5.20

8 / 10

IMO-ProofBench Advanced

Thinking Enabled

4.80

14 / 19

FrontierMath

Standard Mode

4.10

41 / 60

IMO 2025

Standard Mode

3.30

6 / 9

IMO 2025

Thinking Enabled

5 / 9

FrontierMath - Tier 4

Standard Mode

72 / 80

Writing and Creative Capabilities

1 evaluations

Benchmark / mode

Score

Rank/total

Creative Writing

Standard Mode

1480.30

49 / 99

AI Agent - Tool Usage

4 evaluations

Benchmark / mode

Score

Rank/total

OSWorld-Verified

Thinking EnabledTools

42.20

24 / 26

Terminal-Bench

Standard ModeTools

26 / 35

Terminal-Bench

Thinking EnabledTools

35.50

18 / 35

Terminal-Bench

DeepTools

41.30

10 / 35

Multimodal Understanding

1 evaluations

Benchmark / mode

Score

Rank/total

MMMU

Standard Mode

76.50

16 / 28

Common Sense Reasoning

1 evaluations

Benchmark / mode

Score

Rank/total

SimpleBench

Thinking Enabled

45.50

58 / 92

Agent Level Benchmark

4 evaluations

Benchmark / mode

Score

Rank/total

τ²-Bench - Telecom

Thinking EnabledTools

29 / 35

Aider-Polyglot

Standard Mode

56.40

26 / 59

Aider-Polyglot

32K

61.30

20 / 59

τ²-Bench

Standard ModeTools

35 / 44

Instruction Following

1 evaluations

Benchmark / mode

Score

Rank/total

IF Bench

Thinking EnabledTools

28 / 35

Productivity Knowledge

1 evaluations

Benchmark / mode

Score

Rank/total

GDPval-AA

Thinking Enabled

19 / 21

Long Context

1 evaluations

Benchmark / mode

Score

Rank/total

AA-LCR

Thinking Enabled

22 / 28

Claw-style Agent Evaluation

2 evaluations

Benchmark / mode

Score

Rank/total

Pinch Bench

Thinking EnabledTools

80.50

23 / 38

Claw Bench

Thinking EnabledTools

77.80

23 / 29

Compare with other models

Sources

artificialanalysis.aiartificialanalysis.ai