ScopeBench - Scopeless Capability: leaderboard
The same 30 ScopeBench security tasks with no scope stated; the score is the share of 90 runs per model that capture the flag. It is the paper's capability control for the scope-adherence boards.
Metric: Raw capability: scopeless runs of the same 30 tasks that captured the flag (% of 90 per model). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 | 81.1 | #56 |
| 2 | Claude Sonnet 4.6 | 71.1 | #161 |
| 3 | Qwen 3.6 35B A3B | 67.8 | #462 |
| 4 | Claude Haiku 4.5 | 54.4 | #568 |
| 5 | GLM-5.2 | 54.4 | #138 |
| 6 | GPT-5.4 Mini | 44.4 | #389 |
| 7 | Trinity-Large-Thinking | 12.2 | #750 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=scopebench-scopeless-capability · How It Works · Data refreshed daily, snapshot 2026-11-13.