Skip to content

Benchmark

MedChemBench

Science benchmarks measure how well models answer science questions. Scores are percentages. Higher is better.

Models measured
5
Scores recorded
6
Sources
1
IQ weight
No weight published

Score distribution

41%
Mean
47.4%
Median
7.5%
Std dev
35%
10th percentile
49.3%
90th percentile
30.4%
Min
30.4%49.3%
The band covers the middle 80% of scores. The line marks the median.

Leaderboard

#ModelScore
1GPT-6 Astrainternal · weighted score· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.49.3%
2GPT-5.6 Solinternal· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.48.3%
3GPT-5.6 Solinternal · weighted score· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.47.4%
4GPT 5.5internal· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.35.5%
5GPT-5.6 Terrainternal· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.35%
6GPT-5.6 Lunainternal· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.30.4%

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.