Agent Security League - Security Correctness: leaderboard
Agent Security League measures security correctness for coding-agent/model pairs. Scores are retained from each framework run, with date and raw agent identity; independent per-metric maxima are not synthesized across agents. Existing downstream harness selection remains separate from extraction.
Metric: Security Correctness (%). Source: www.endorlabs.com. Saturation forecast: Around August 2028. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5.1 | 37.4 | #9 |
| 2 | GPT-6 Astra | 34.6 | #6 |
| 3 | GPT-6.1 Sol | 34.1 | #7 |
| 4 | Claude Opus 5.5 | 33.5 | #3 |
| 5 | Claude Opus 5 | 32.4 | #12 |
| 6 | GPT-6 Sol | 25.1 | #14 |
| 7 | GPT-5.5 | 24 | #31 |
| 8 | GPT-5.6 Sol | 23.5 | #19 |
| 9 | GPT-5.4 | 21.8 | #79 |
| 10 | Claude Sonnet 5 | 19.6 | #57 |
| 11 | Claude Fable 5 | 19 | #16 |
| 12 | Gemini 3.5 Flash | 17.9 | #60 |
| 13 | Claude Opus 4.7 | 16.8 | #49 |
| 14 | Gemini 3.1 Pro (Preview) | 15.1 | #58 |
| 15 | Claude Opus 4.8 | 14.5 | #37 |
No result here: #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=agent-security-league-security-correctness · How It Works · Data refreshed daily, snapshot 2026-10-20.