Kaggle FACTS Parametric: leaderboard
Short closed-book questions with one atomic factual answer documented on Wikipedia, kept only where five open-weight models all failed and human annotators confirmed the answer. Models reply with no search or other tools, and Gemini 2.5 Pro grades each reply correct, incorrect, not attempted or unknown. The board reports accuracy over public and private halves of 2,104 items, and higher is better.
Metric: Score (%). Source: www.kaggle.com. Saturation forecast: Around January 2027. 44 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 83.08 | #6 |
| 2 | GPT-6.1 Sol | 82.11 | #7 |
| 3 | Gemini 3.1 Pro (Preview) | 78.96 | #58 |
| 4 | Gemini 3.7 Flash | 78.72 | #23 |
| 5 | GPT-5.5 | 78.04 | #31 |
| 6 | Gemini 3.8 Flash | 77.34 | #21 |
| 7 | GPT-6 Sol | 75.59 | #14 |
| 8 | Gemini 3.5 Flash | 73.46 | #60 |
| 9 | Gemini 3.6 Flash | 73.35 | #51 |
| 10 | Gemini 3 Flash (Preview) | 72.26 | #82 |
| 11 | Gemini 4 Argon | 71.55 | #8 |
| 12 | Gemini 2.5 Pro | 63.21 | #152 |
| 13 | Grok 4 | 58.56 | #181 |
| 14 | Claude Opus 5.5 | 57.87 | #3 |
| 15 | O3 | 57.08 | #124 |
No result here: #9 Claude Fable 5.1, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=kaggle-facts-parametric · How It Works · Data refreshed daily, snapshot 2026-10-20.