Sonar LLM Leaderboard - Java - Vulnerability Density: leaderboard
SonarQube vulnerabilities (security issues) per 1,000 lines of generated Java code on Sonar's LLM Leaderboard, which has each model solve a fixed set of Java programming assignments and analyses the generated code with SonarQube; lower is better.
Metric: Vulnerabilities per 1,000 lines of code (lower is better). Source: www.sonarsource.com. Saturation forecast: Around November 2028. 74 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (Medium) | 0.04 | #83 (GPT-5.4) |
| 2 | GPT-5.4 Nano (Medium) | 0.04 | #359 (GPT-5.4 Nano) |
| 3 | GPT-5.4 (High) | 0.05 | #83 (GPT-5.4) |
| 4 | GPT-5.2 (High) | 0.05 | #115 (GPT-5.2) |
| 5 | GPT-5.5 (Medium) | 0.06 | #33 (GPT-5.5) |
| 6 | GPT-5.4 Mini (Medium) | 0.06 | #228 (GPT-5.4 Mini) |
| 7 | GPT-5.5 (Low) | 0.06 | #33 (GPT-5.5) |
| 8 | GPT-5.3 Codex (High) | 0.07 | #72 (GPT-5.3 Codex) |
| 9 | GPT-5.1 (High) | 0.09 | #145 (GPT-5.1) |
| 10 | GPT-5.1 (Medium) | 0.11 | #145 (GPT-5.1) |
| 11 | GPT-5.1 Codex (High) | 0.11 | #168 (GPT-5.1 Codex) |
| 12 | Claude Opus 4.5 (High) | 0.13 | #87 (Claude Opus 4.5) |
| 13 | Claude Opus 4.6 (High) | 0.14 | #67 (Claude Opus 4.6) |
| 14 | GPT-6 Astra (Medium) | 0.18 | #6 (GPT-6 Astra) |
| 15 | Gemini 3 Pro (High) | 0.18 | #82 (Gemini 3 Pro) |
No result here: #10 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=sonar-llm-leaderboard-java-vulnerability-density · How It Works · Data refreshed daily, snapshot 2026-11-05.