Skip to content

Benchmark

Investment Banking Modeling Tasks

Scores are percentages. Higher is better.

Models measured
7
Scores recorded
9
Sources
1
IQ weight
0.58

Score distribution

78.6%
Mean
79.3%
Median
8.2%
Std dev
68.4%
10th percentile
88.5%
90th percentile
68.4%
Min
68.4%88.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.58 for chat models, measured over 7 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT-5.5 Prointernal · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.6%
2GPT 5.5internal · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.88.5%
3GPT 5.4OpenAI internal benchmark· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.87.3%
4GPT-5.4 ProOpenAI internal benchmark· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.83.6%
5GPT-5.3 CodexOpenAI internal benchmark· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.79.3%
6GPT-5.2 Prointernal spreadsheet tasks · ChatGPT heavy· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.71.7%
7GPT-5.2 ProOpenAI internal benchmark· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.71.7%
8GPT 5.2internal spreadsheet tasks · ChatGPT heavy· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.68.4%
9GPT 5.2OpenAI internal benchmark· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.68.4%

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.