HardSecBench - Single Attempt@1 - Security: leaderboard
Metric: Requirement Pass Rate (%). Source: arxiv.org. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 42.51 |
| 2 | GPT-5.1 (Medium) | 40.02 |
| 3 | Gemini 2.5 Pro | 39.29 |
| 4 | GPT-5 (Medium) | 39.03 |
| 5 | Claude Opus 4.5 | 37.45 |
| 6 | Llama 4 Maverick Instruct | 35.87 |
| 7 | GLM-4.6 | 35.67 |
| 8 | O1 | 35.57 |
| 9 | GPT-OSS-120B | 34.07 |
| 10 | Kimi K2 0905 | 33.06 |
| 11 | Qwen 3 Coder 480B A35B | 32.74 |
| 12 | Claude Sonnet 4.5 | 32.31 |
| 13 | DeepSeek V3.2 | 32.11 |
| 14 | Gemini 2.5 Flash | 32.07 |
| 15 | Llama 4 Scout Instruct | 30.82 |
Interactive version: theaggregate.ai/benchmark?slug=hardsecbench-single-attempt-1-security · How It Works · Data refreshed daily, snapshot 2026-09-19.