Capability
Weighted BlueBench score across incident response, threat hunting, detection engineering, and malware analysis. Higher means more capable.
26 models · Updated Oct 2026
Methodology Score
Models
- 1Claude Opus 5.5
- 2GPT-6 Astra
- 3Gemini 4 Argon
- 4Grok 4.6
- 5Claude Sonnet 5.5
- 6GPT-5.6 Sol
- 7GPT-6.1 Sol
- 8Claude Opus 4.8
- 9DeepSeek V4.1 Flash
- 10GPT-6 Sol
- 11GLM-5.3
- 12GLM-5.3 Flash
- 13GPT-6.1 Terra
- 14GPT-5.6 Terra
- 15Claude Haiku 5.5
- 16Kimi K3
- 17GLM-5.2
- 18GPT-6 Luna
- 19Gemini 3.6 Flash
- 20GPT-5.6 Luna
- 21DeepSeek V4 Flash
- 22MiniMax M3
- 23DeepSeek V4 Pro
- 24Claude Opus 5
- 25Gemini 3.1 Pro
- 26Claude Sonnet 5
| Rank | Models (26) | Status | ||||||
|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 | 83.2 | 87.8 | 76.1 | 83.8 | 89.4 | $3.54 | |
| 2 | GPT-6 Astra | 80.3 | 88.0 | 62.8 | 87.3 | 88.8 | $9.60 | |
| 3 | Gemini 4 Argon | 79.9 | 86.8 | 74.5 | 80.7 | 72.5 | $6.76 | |
| 4 | Grok 4.6 | 79.2 | 80.8 | 72.6 | 82.6 | 83.8 | $3.64 | |
| 5 | Claude Sonnet 5.5 | 77.5 | 83.9 | 66.9 | 78.1 | 87.8 | $1.01 | |
| 6 | GPT-5.6 Sol | 77.2 | 82.6 | 62.7 | 82.8 | 87.5 | $2.89 | |
| 7 | GPT-6.1 Sol | 76.7 | 86.4 | 61.7 | 78.5 | 87.5 | $1.30 | |
| 8 | Claude Opus 4.8 | 74.9 | 68.4 | 68.6 | 85.3 | 81.9 | $1.56 | |
| 9 | DeepSeek V4.1 Flash | 71.5 | 60.9 | 68.5 | 81.7 | 81.9 | $0.17 | |
| 10 | GPT-6 Sol | 71.1 | 77.9 | 56.2 | 78.4 | 73.8 | $1.74 | |
| 11 | GLM-5.3 | 70.9 | 77.2 | 68.2 | 64.1 | 80.3 | $3.27 | |
| 12 | GLM-5.3 Flash | 70.6 | 72.6 | 63.2 | 74.7 | 74.7 | $0.11 | |
| 13 | GPT-6.1 Terra | 70.5 | 78.2 | 59.6 | 72.5 | 74.1 | $0.16 | |
| 14 | GPT-5.6 Terra | 69.9 | 69.1 | 62.1 | 77.4 | 73.1 | $1.01 | |
| 15 | Claude Haiku 5.5 | 68.5 | 65.5 | 59.9 | 80.5 | 67.8 | $0.59 | |
| 16 | Kimi K3 | 67.3 | 72.3 | 64.3 | 62.0 | 77.2 | $2.62 | |
| 17 | GLM-5.2 | 65.5 | 62.8 | 52.4 | 75.9 | 81.9 | $1.03 | |
| 18 | GPT-6 Luna | 64.1 | 71.1 | 51.0 | 69.9 | 65.3 | $0.26 | |
| 19 | Gemini 3.6 Flash | 62.7 | 54.5 | 53.0 | 81.1 | 61.6 | $3.01 | |
| 20 | GPT-5.6 Luna | 62.4 | 61.8 | 52.4 | 67.5 | 79.1 | $0.58 | |
| 21 | DeepSeek V4 Flash | 58.6 | 57.7 | 51.9 | 64.8 | 62.8 | $0.17 | |
| 22 | MiniMax M3 | 56.4 | 54.3 | 53.8 | 56.8 | 69.1 | $0.90 | |
| 23 | DeepSeek V4 Pro | 53.6 | 51.6 | 41.3 | 65.6 | 60.0 | $0.91 | |
| 24 | Claude Opus 5 | 50.5 | 37.8 | 36.1 | 76.3 | 54.7 | $4.44 | Cyber refused |
| 25 | Gemini 3.1 Pro | 45.3 | 32.6 | 43.2 | 55.8 | 58.3 | $4.63 | |
| 26 | Claude Sonnet 5 | 44.8 | 29.2 | 23.8 | 67.3 | 86.9 | $1.77 | Cyber refused |
Cyber refused: the model refused a material share of tasks on cybersecurity grounds. Refused tasks stay in its scores.
AI for the blue team.
Run Cotool's harness in your environment to get real security work done