XOR Unpatched - Defend (Full Fix): leaderboard
XOR Unpatched, defender role: vulnerabilities XOR researchers discovered and disclosed in a confidential inference stack, with no public fix, so no model has seen them. A full fix must close the flaw and keep legitimate use working; partial credit is not counted here. Systems are one harness plus model at a recorded reasoning effort, unranked by XOR because runs are few; percent of graded runs with a full fix, higher is better.
Metric: Full Fix Rate (%). Source: www.xor.tech. Saturation forecast: Around June 2028. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (xHigh) | 35.3 | #6 (GPT-6 Astra) |
| 2 | Claude Fable 5.1 (High) | 28.6 | #10 (Claude Fable 5.1) |
| 3 | GPT-5.6 Sol (xHigh) | 16.7 | #22 (GPT-5.6 Sol) |
| 4 | Claude Opus 5 (Max) | 12.9 | #16 (Claude Opus 5) |
| 5 | Qwen 3.8 Max (xHigh) | 4.2 | #34 (Qwen 3.8 Max) |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=xor-unpatched-defend-full-fix · How It Works · Data refreshed daily, snapshot 2026-11-05.