Skip to content

Benchmark

OfficeQA

Scores are percentages. Higher is better.

Models measured
13
Scores recorded
16
Sources
2
IQ weight
1.31

Score distribution

74.2%
Mean
76.9%
Median
6.8%
Std dev
65.1%
10th percentile
83.4%
90th percentile
63%
Min
63%86.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.31 for chat models, measured over 13 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 4.7exact-match grading · 0% allowable relative error· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.3%
2Claude Fable 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim83.4%
3Claude Fable 5.1Opus 5.5 System Card capability section · same Opus 5.5 card setup· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.2%
4Claude Opus 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim78.9%
5Claude Opus 5.5Opus 5.5 System Card capability section · extracted-text sandbox + code · max · 5 runs· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.78.9%
6Claude Opus 5Opus 5.5 System Card capability section · same Opus 5.5 card setup· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.78.1%
7Claude Sonnet 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim76.9%
8Claude Sonnet 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · preselected extracted text + code · max · 5 runs· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.76.9%
9Claude Haiku 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim73.5%
10Claude Haiku 5.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · preselected extracted text + code · max · 5 runs· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.5%
11Claude Sonnet 5extracted-text sandbox + code · exact match · max effort · 5 trials· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.3%
12Claude Sonnet 4.6extracted-text sandbox + code · max effort · Sonnet 5 card rerun· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.68.7%
13GPT 5.4OpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.68.1%
14GPT-5.3 CodexOpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.65.1%
15GPT 5.2OpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.63.1%
16Claude Haiku 4.5adaptive thinking · max effort · default sampling · 5-trial mean · context ≤1M unless noted · preselected extracted text + code · fixed thinking budget · 5 runs· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.63%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats