Cybersecurity Evaluation on Cybersecurity Benchmarks CTI-MCQ, CyberMetric, SecEval (test)
73.79CTI-MCQ ScoreGooseReason-Cyber-4B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GooseReason-Cyber-4BModel Size=4B, Training Data=GooseReason-Cyber, RL Steps=1002026.01 | 73.79 | 92.05 | 71.14 | 78.99 | |
| Llama-Primus-InstructModel Size=8B, Domain=Cybersecurity-specialized2026.01 | 66.6 | 86.4 | 49.43 | 67.48 | |
| Llama-Primus-MergedModel Size=8B, Domain=Cybersecurity-specialized2026.01 | 66.56 | 86.6 | 50.62 | 67.93 | |
| Llama-3.1-8B-InstructModel Size=8B, Domain=General2026.01 | 64.2 | 85.6 | 49.66 | 66.49 | |
| Qwen3-4B-InstructModel Size=4B, Domain=General Reasoning2026.01 | 63.44 | 89.78 | 70.44 | 74.55 |