DataLearner logo

Opus 4.5 Benchmark Details

Opus 4.5 currently shows benchmark results led by MMLU Pro (2 / 126, score 90), SWE-bench Verified (8 / 109, score 80.90), Terminal Bench Hard (1 / 13, score 44). 3 source links are attached for reference.

Benchmark Results

Opus 4.5

Benchmark Results

Thinking
Tool usage

General Knowledge

10 evaluations
Benchmark / mode
Score
Rank/total
MMLU Pro
Extended
90
2 / 126
GPQA Diamond
Extended
87
40 / 180
ARC-AGI
Extended
80
21 / 65
LiveBench
64K
75.96
11 / 115
LiveBench
Thinking Level · Low
55.77
80 / 115
LiveBench
Thinking Level · Medium
59.10
72 / 115
LiveBench
Thinking Level · High
58.59
74 / 115
HLE
Extended
30.80
77 / 164
HLE
ExtendedTools
43.20
43 / 164
ARC-AGI-2
Extended
37.60
26 / 59

Coding and Software Engineer

2 evaluations
Benchmark / mode
Score
Rank/total
LiveCodeBench
ExtendedTools
87
12 / 120
SWE-bench Verified
ExtendedTools
80.90
8 / 109

Multimodal Understanding

1 evaluations
Benchmark / mode
Score
Rank/total
MMMU
Extended
80.70
10 / 28

常识推理

1 evaluations
Benchmark / mode
Score
Rank/total
Simple Bench
Extended
62
12 / 63

Agent Level Benchmark

3 evaluations
Benchmark / mode
Score
Rank/total
τ²-Bench - Telecom
ExtendedTools
90.70
21 / 35
τ²-Bench
ExtendedTools
81.99
13 / 40
Terminal Bench Hard
ExtendedTools
44
1 / 13

Math and Reasoning

6 evaluations
Benchmark / mode
Score
Rank/total
AIME 2026
Extended
93.30
6 / 15
FrontierMath
Extended
20.70
17 / 60
FrontierMath - Tier 4
Standard Mode
4.20
40 / 80
FrontierMath - Tier 4
16K
2.10
56 / 80
FrontierMath - Tier 4
32K
4.20
40 / 80
FrontierMath - Tier 4
Extended
4.20
40 / 80

Instruction Following

1 evaluations
Benchmark / mode
Score
Rank/total
IF Bench
ExtendedTools
58
20 / 29

AI Agent - Tool Usage

2 evaluations
Benchmark / mode
Score
Rank/total
MCP-Atlas
Thinking Level · HighTools
69.80
14 / 25
Terminal Bench 2.0
ExtendedTools
59.30
20 / 46

Claw-style Agent Evaluation

2 evaluations
Benchmark / mode
Score
Rank/total
Claw Bench
ExtendedTools
91.50
7 / 29
Pinch Bench
ExtendedTools
87.20
8 / 37

Long Context

1 evaluations
Benchmark / mode
Score
Rank/total
LongBench v2
Standard Mode
64.40
1 / 11
Compare with other models

Sources

pinchbench.compinchbench.comanthropic.comanthropic.comepoch.aiepoch.ai