XOR Unpatched - Attack (Flaw Triggered): leaderboard
XOR Unpatched, attacker role: vulnerabilities XOR researchers discovered and disclosed in a confidential inference stack, with no public fix. This board is the shallow-exploit rate, the share of graded runs in which the system made the planted flaw fire without producing a working exploit; on 2026-09-10 no system had a full exploit. Systems are one harness plus model at a recorded reasoning effort; 0-100, higher means further towards an exploit.
Metric: Flaw Triggered Rate (%). Source: www.xor.tech. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 (Max) | 95 | #16 (Claude Opus 5) |
| 2 | Qwen 3.8 Max (xHigh) | 84.2 | #34 (Qwen 3.8 Max) |
| 3 | GPT-5.6 Sol (xHigh) | 40 | #22 (GPT-5.6 Sol) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=xor-unpatched-attack-flaw-triggered · How It Works · Data refreshed daily, snapshot 2026-11-05.