HardSecBench - Refinement@1 (GPT-5.1 Collaborator) - Security: leaderboard
Measures functional and security requirement pass rates on 924 hardware/firmware tasks. Single-attempt permits up to three compiler-only repairs. Refinement uses GPT-5.1 functional-only collaborator feedback for up to five iterations. Pass@k averages atomic requirements, not whole-task success. Security hints are absent in this main evaluation. Model effort and dated labels are retained. This series reports HardSecBench - Refinement@1 (GPT-5.1 Collaborator) - Security using Requirement Pass Rate (%).
Metric: Requirement Pass Rate (%). Source: arxiv.org. Saturation forecast: Around 2034. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 46.11 | #71 |
| 2 | GPT-5.1 (Medium) | 45.76 | #145 (GPT-5.1) |
| 3 | Gemini 2.5 Pro | 44.56 | #159 |
| 4 | Llama 4 Maverick Instruct | 44.38 | #467 |
| 5 | GPT-5 (Medium) | 42.96 | #105 (GPT-5) |
| 6 | Claude Opus 4.5 | 42.91 | #87 |
| 7 | GLM-4.6 | 42.89 | #265 |
| 8 | DeepSeek V3.2 | 42.42 | #210 |
| 9 | O1 | 41.66 | #190 |
| 10 | GPT-OSS-120B | 40.73 | #346 |
| 11 | Llama 4 Scout Instruct | 40.66 | #547 |
| 12 | Qwen 3 14B | 40.57 | #542 |
| 13 | Qwen 3 Coder 480B A35B | 39.74 | #286 |
| 14 | Kimi K2 0905 | 39.09 | #292 |
| 15 | Claude Sonnet 4.5 | 38.68 | #148 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hardsecbench-refinement-1-gpt-5-1-collaborator-security · How It Works · Data refreshed daily, snapshot 2026-11-05.