Spiral Bench (EQ-Bench): leaderboard
Tests whether models resist sycophancy and avoid reinforcing delusional thinking. 30 x 20-turn simulated chats judged by an ensemble of frontier models.
Metric: Delusion Score (0-100). Source: eqbench.com. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 70.3 | #142 |
| 2 | GPT-5.2 | 70.2 | #110 |
| 3 | O3 | 63.4 | #124 |
| 4 | GPT-OSS-120B | 57.5 | #341 |
| 5 | GPT-5 | 57.4 | #98 |
| 6 | Kimi K2 | 57.2 | #243 |
| 7 | O4 Mini | 53.6 | #180 |
| 8 | Claude 3.5 Sonnet | 42.6 | #350 |
| 9 | Gemini 3 Pro (Preview) | 38.7 | #66 |
| 10 | Llama 4 Maverick | 38.2 | #464 |
| 11 | Grok 4 Fast | 37.5 | #254 |
| 12 | Gemini 2.5 Flash | 36.5 | #245 |
| 13 | Claude Sonnet 4 | 32.1 | #199 |
| 14 | GLM-4.6 | 30.8 | #252 |
| 15 | Gemini 2.5 Pro | 28.3 | #152 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=spiral-bench-eq-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.