Vulnerability Discovery on OSS benchmark (371 held-out cases)
412High/Critical Qualified Findings RateEVOHUNT (GLM-5.1)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| EVOHUNT (GLM-5.1)Backbone=GLM-5.1, Condition=Exp Evolve2026.06 | 412 | 42 | 31 | 8 | 3 | 370 | |
| Exp Empty GLMBackbone=GLM-5.1, Condition=Empty Playbook2026.06 | 247 | 36 | 18 | 0 | 18 | 211 | |
| Exp OCS GPTBackbone=GPT-5.4-xhigh, Condition=OCS Product-style reference2026.06 | 194 | 34 | 13 | 4 | 17 | 160 | |
| EVOHUNT (GPT-5.4)Backbone=GPT-5.4-xhigh, Condition=Exp Evolve2026.06 | 120 | 23 | 23 | 0 | 0 | 97 | |
| Exp Empty GPTBackbone=GPT-5.4-xhigh, Condition=Empty Playbook2026.06 | 34 | 6 | 4 | 2 | 0 | 28 |