For Your Eyes Only - Constrained: leaderboard
The constrained setting of For Your Eyes Only: explicit signalling is forbidden and a naive Inspector model rejects descriptions that a casual reader would spot, so only implicit coordination counts. Double-pass success rate, higher is better. Source: arxiv.org/abs/2609.19504.
Metric: Double-Pass Success Rate (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 96.33 | #64 |
| 2 | Gemini 3 Flash | 65 | #103 |
| 3 | GPT-OSS-120B | 53.67 | #346 |
| 4 | DeepSeek V4 Flash | 39 | #155 |
| 5 | Qwen 3.6 Plus | 39 | #124 |
| 6 | DeepSeek V4 Pro | 24 | #112 |
| 7 | Qwen 3.6 35B A3B | 8.67 | #238 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=for-your-eyes-only-constrained · How It Works · Data refreshed daily, snapshot 2026-11-05.