Four Bridges - Deceptive share: leaderboard
Four Bridges Table 15:49 baseline games per informed model, authored-message reading; three-judge majority classifies deception, mixed behavior separate. Deceptive share is a behavioral safety diagnostic, not game score or task competence. Exact API generation identities in Table 8; source effort unspecified. Printed percentages preserved; Wilson intervals available in primary source. Metric: Deceptive share (%).
Metric: Deceptive share (%). Source: arxiv.org. Saturation forecast: Not forecast. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Terra | 6 | #40 |
| 2 | GPT-5.6 Luna | 49 | #79 |
| 3 | Claude Sonnet 5 | 59 | #65 |
| 4 | GLM-5.2 | 78 | #80 |
| 5 | GPT-5.6 Sol | 90 | #22 |
| 6 | Claude Opus 5 | 100 | #16 |
| 7 | Claude Fable 5 | 100 | #20 |
| 8 | Grok 4.6 | 100 | #29 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=four-bridges-deceptive-share · How It Works · Data refreshed daily, snapshot 2026-11-05.