AIRTBench: leaderboard
AI red teaming benchmark evaluating language models' ability to autonomously discover and exploit AI/ML security vulnerabilities across 70 security challenges.
Metric: Success Rate (%). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet (20250219) | 46.86 | #201 |
| 2 | GPT-4.5 (Preview) | 36.89 | #244 |
| 3 | Gemini 2.5 Pro (Preview 05-06) | 34.29 | #147 |
| 4 | O3 Mini | 28.43 | #272 |
| 5 | DeepSeek R1 | 26.86 | #251 |
| 6 | Gemini 2.5 Flash (Preview 04-17) | 26.43 | #237 |
| 7 | GPT-4o | 20.29 | #343 |
| 8 | Gemini 2.0 Flash | 16.86 | #342 |
| 9 | Gemini 1.5 Pro | 15.14 | #445 |
| 10 | Llama 4 Scout Instruct | 1 | #539 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=airtbench · How It Works · Data refreshed daily, snapshot 2026-10-20.