Anthropic

Claude Opus 4.8

Comprehensive benchmark performance across 11 evaluation categories

Composite Score

0.0/100

Rank

Token Benchmark

68.3

Lower burn, higher score

Total Tokens

134.7k

~7.1k/test

Category Radar

The full radar chart is shown on wider screens. On mobile, the category breakdown below provides the same values in a readable stacked layout.

Historical Composite Score

Category Breakdown

Bug Introduction Rate

100.0

1 tests1.1x weight

Instruction Following

100.0

1 tests0.8x weight

Coding Tasks

99.0

2 tests1.2x weight

Feature Implementation

96.7

3 tests1.0x weight

Security Awareness

94.5

2 tests1.1x weight

Performance & Efficiency

94.0

3 tests1.0x weight

Code Quality

93.5

2 tests0.9x weight

Bug Fixes

90.0

1 tests1.2x weight

Long Reasoning

77.8

2 tests1.0x weight

#10

Code Thoroughness

74.0

2 tests0.9x weight

#11

Token Efficiency

68.3

19 tests1.0x weight

#	Category	Score	Tests	Weight
1	Bug Introduction Rate	100.0	1	1.1x
2	Instruction Following	100.0	1	0.8x
3	Coding Tasks	99.0	2	1.2x
4	Feature Implementation	96.7	3	1.0x
5	Security Awareness	94.5	2	1.1x
6	Performance & Efficiency	94.0	3	1.0x
7	Code Quality	93.5	2	0.9x
8	Bug Fixes	90.0	1	1.2x
9	Long Reasoning	77.8	2	1.0x
10	Code Thoroughness	74.0	2	0.9x
11	Token Efficiency	68.3	19	1.0x

Individual Test Results

Token Efficiency68.3

Token Efficiency is computed from every successful task in the run. The model with the lowest average token burn receives 100, and heavier token usage is penalized proportionally.

Avg Tokens/Test

7.1k

Total Tokens

134.7k

Long Reasoning77.8

Mathematical Proof

4.5k tok25.3s78.7

Legal Reasoning Chain

6.2k tok52.1s77.0

Coding Tasks99.0

Graph Algorithm Implementation

3.9k tok17.4s98.0