CresOWLve: leaderboard
Metric: Accuracy (%) judged by GPT-4o against the reference answer, its explanation and accepted alternatives, English version, on the 2,061 creative questions of CresOWLve (expert-written What? Where? When? puzzles that need several real-world facts joined by a creative leap), zero-shot; non-thinking models with chain-of-thought prompting, thinking models at the stated effort; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 76.03 |
| 2 | Gemini 3.1 Pro (Preview) (Medium) | 72.97 |
| 3 | Gemini 3.1 Pro (Preview) (Low) | 67.54 |
| 4 | GPT-5.4 (Medium) | 53.47 |
| 5 | Gemini 3 Flash (Medium) | 52.16 |
| 6 | Gemini 3 Flash (Minimal) | 41.1 |
| 7 | GLM-5 | 37.17 |
| 8 | Qwen 3.5 397B A17B | 31.25 |
| 9 | DeepSeek V3.2 (Thinking) | 25.47 |
| 10 | GPT-4.1 | 24.99 |
| 11 | GPT-5.4 (Non-reasoning) | 20.96 |
| 12 | Qwen 3 235B A22B 2507 Instruct | 20.82 |
| 13 | Qwen 3 235B A22B 2507 (Thinking) | 20.82 |
| 14 | Mistral-Large-3-675B-Instruct-2512 | 17.61 |
| 15 | GPT-4.1 Mini | 13.78 |
Interactive version: theaggregate.ai/benchmark?slug=cresowlve · How It Works · Data refreshed daily, snapshot 2026-10-07.