Aikido CVE Rediscovery: leaderboard
Rediscovery of 26 known real-world CVEs (XSS, unsafe deserialization, access control, command injection, SQL injection, race conditions, DoS) in their original codebases, with the fix reverted and no hint of where the flaw is; mean recall over three independent runs per model.
Metric: Recall (%). Source: www.aikido.dev. Saturation forecast: Around August 2027. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Terra (Max) | 83.3 | #41 (GPT-5.6 Terra) |
| 2 | GPT-5.6 Sol (High) | 78.2 | #19 (GPT-5.6 Sol) |
| 3 | GPT-5.6 Luna (Max) | 74.4 | #68 (GPT-5.6 Luna) |
| 4 | GPT-5.5 (xHigh) | 73.1 | #31 (GPT-5.5) |
| 5 | GPT-5.6 Sol (Max) | 73.1 | #19 (GPT-5.6 Sol) |
| 6 | GPT-5.6 Terra (High) | 69.2 | #41 (GPT-5.6 Terra) |
| 7 | GPT-5.6 Luna (High) | 65.4 | #68 (GPT-5.6 Luna) |
| 8 | GPT-5.4 Mini (High) | 64.1 | #212 (GPT-5.4 Mini) |
| 9 | GPT-5.4 Mini (xHigh) | 64.1 | #212 (GPT-5.4 Mini) |
| 10 | GPT-5.5 (High) | 62.8 | #31 (GPT-5.5) |
| 11 | Grok 4.5 (High) | 62.8 | #48 (Grok 4.5) |
| 12 | GPT-5.4 Nano (xHigh) | 57.7 | #328 (GPT-5.4 Nano) |
| 13 | Claude Opus 4.8 (xHigh) | 55.1 | #37 (Claude Opus 4.8) |
| 14 | GPT-5.4 Nano (High) | 50 | #328 (GPT-5.4 Nano) |
| 15 | Gemini 3.1 Pro (Preview) (High) | 47.4 | #58 (Gemini 3.1 Pro (Preview)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=aikido-cve-rediscovery · How It Works · Data refreshed daily, snapshot 2026-10-20.