RealVuln - F1: leaderboard
RealVuln (2026) benchmarks security scanners on 26 intentionally vulnerable Python repositories with 796 hand-labeled entries, including false-positive traps; LLMs scan in an agentic loop with file, search and shell tools under one shared prompt. This board is the strict F1 score of the findings in percent; higher is better.
Metric: Strict F1 score of vulnerability findings against 796 hand-labeled entries in 26 intentionally vulnerable Python repositories, agentic scanning; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 60.9 | #173 | |
| 2 | Gemini 3.1 Pro (Preview) | 60.1 | #99 | |
| 3 | Claude Opus 4.6 | 57.8 | #113 | |
| 4 | GLM-5 | 55.5 | #269 | |
| 5 | Kimi K2.5 | 54.3 | #266 | |
| 6 | MiniMax-M2.7 | 48.7 | #539 | |
| 7 | Claude Haiku 4.5 | 47.9 | #583 | |
| 8 | Qwen 3.5 397B A17B | 46 | #264 | |
| 9 | Grok 4.20 | Reasoning | 41 | #301 |
| 10 | Grok 3 | 31.9 | #611 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=realvuln-f1 · How It Works · Data refreshed daily, snapshot 2026-11-21.