Skip to content

Benchmark

Terminal-Bench-Science 0.1 Evaluation Results

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages.

Models measured
14
Scores recorded
14
Sources
1
IQ weight
2.42

Score distribution

17.9%
Mean
10.5%
Median
20.1%
Std dev
3.8%
10th percentile
63.3%
90th percentile
3.3%
Min
3.3%68.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.42 for chat models, measured over 14 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1GPT-6 Astrav0-1-eval· Harbor Hub Leaderboards68.1%
2Claude Opus 5.5v0-1-eval· Harbor Hub Leaderboards63.3%
3GPT-5.6 Solv0-1-eval· Harbor Hub Leaderboards22.4%
4DeepSeek V4.1 Flashv0-1-eval· Harbor Hub Leaderboards15.7%
5Grok 4.7v0-1-eval· Harbor Hub Leaderboards14.3%
6Gemini 3.8 Flashv0-1-eval· Harbor Hub Leaderboards12.4%
7Claude Opus 4.8v0-1-eval· Harbor Hub Leaderboards10.5%
8GPT-5.6 Terrav0-1-eval· Harbor Hub Leaderboards8.6%
9GLM 5.3v0-1-eval· Harbor Hub Leaderboards8.1%
10Grok 4.6v0-1-eval· Harbor Hub Leaderboards7.1%
11Kimi K3v0-1-eval· Harbor Hub Leaderboards7.1%
12Gemini 3.7 Flashv0-1-eval· Harbor Hub Leaderboards5.7%
13DeepSeek V4 Prov0-1-eval· Harbor Hub Leaderboards3.8%
14GPT-5.6 Lunav0-1-eval· Harbor Hub Leaderboards3.3%

Tokens per task, input and output together, as measured by the source. Comparable within this benchmark only.