Skip to main content

BlueBench results by model

How each frontier and open-weight model performs on defensive security work: overall score, per-track and per-benchmark results, cost per task, and reliability.

26 models ยท Updated Oct 2026
Methodology

All models

Ranked by overall BlueBench score. Open a model for its full results.

ModelRankScoreCost / task
#183.2$3.54
#280.3$9.60
#379.9$6.76
#479.2$3.64
#577.5$1.01
#677.2$2.89
#776.7$1.30
#874.9$1.56
#971.5$0.17
#1071.1$1.74
#1170.9$3.27
#1270.6$0.11
#1370.5$0.16
#1469.9$1.01
#1568.5$0.59
#1667.3$2.62
#1765.5$1.03
#1864.1$0.26
#1962.7$3.01
#2062.4$0.58
#2158.6$0.17
#2256.4$0.90
#2353.6$0.91
#2450.5$4.44
#2545.3$4.63
#2644.8$1.77

Open-weight models

Models with published weights that you can host yourself.

ModelRankScoreCost / task
#971.5$0.17
#1170.9$3.27
#1270.6$0.11
#1667.3$2.62
#1765.5$1.03
#2158.6$0.17
#2256.4$0.90
#2353.6$0.91

AI for the blue team.

Run Cotool's harness in your environment to get real security work done

Book a demo