BoxPwnr CTF Bench: leaderboard
BoxPwnr single_loop solver: unweighted mean of published scores over at least three distinct security-lab platforms per model. Models can cover different platform sets; each contribution retains its task counts and pass-at-n protocol. The scores include capability coverage where the platform defines it and are not uniformly completion rates. Solver protocols are separate boards.
Metric: Mean published platform score (%); observed platforms only; single_loop. Source: github.com. Saturation forecast: Around January 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.1 | 55.47 | #97 |
| 2 | GLM-5 | 35.08 | #141 |
| 3 | Kimi K2.5 | 30 | #144 |
| 4 | Qwen 3.5 122B A10B | 22.67 | #173 |
| 5 | Grok 4.1 Fast | 13.33 | #217 |
| 6 | Claude Sonnet 4.5 | 1.8 | #142 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=boxpwnr-ctf-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.