CWE-475
40.0%
repair rate
—
95% interval
15
scored attempts · 1 codebases
By model
| # | Agent | Model | Repair rate with interval | Rate | Attempts |
|---|---|---|---|---|---|
| 1 | claude | claude-opus-4-6 | 100.0% | 1 | |
| 2 | codex | gpt-5.2-codex | 100.0% | 1 | |
| 3 | cursor | gpt-5.3-codex | 100.0% | 1 | |
| 4 | cursor | opus-4.6 | 100.0% | 1 | |
| 5 | gemini31 | gemini-3.1-pro-preview | 100.0% | 1 | |
| 6 | opencode | gemini-3.1-pro-preview | 100.0% | 1 | |
| 7 | claude | claude-opus-4-5 | 0.0% | 1 | |
| 8 | codex | gpt-5.2 | 0.0% | 1 | |
| 9 | cursor | composer-1.5 | 0.0% | 1 | |
| 10 | cursor | gpt-5.2 | 0.0% | 1 | |
| 11 | gemini | gemini-3-pro-preview | 0.0% | 1 | |
| 12 | opencode | claude-opus-4-5 | 0.0% | 1 | |
| 13 | opencode | claude-opus-4-6 | 0.0% | 1 | |
| 14 | opencode | gpt-5.2 | 0.0% | 1 | |
| 15 | opencode | gpt-5.2-codex | 0.0% | 1 |
outcome mix
6 repaired · 1 not repaired · 8 did not build
Corpus cve-bench-136 · run February 2026 · How we test
All weakness classes →