Kaggle Time Traveller's Dilemma: leaderboard
Community benchmark on Kaggle by tiffafoster: 48 prompts asking the model to explain a modern concept to a historical figure using only ideas of that era. A prompt passes only if the reply keeps the required bracketed format, stays within its word range, avoids a list of forbidden modern words and satisfies a judge model; percent of prompts passed, higher is better.
Metric: Scenarios Passed (%). Source: www.kaggle.com. Saturation forecast: Around February 2027. 30 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 81.25 | #64 |
| 2 | Gemini 3.5 Flash | 81.25 | #66 |
| 3 | GPT-5.5 | 79.17 | #33 |
| 4 | Claude Sonnet 4.6 | 77.08 | #94 |
| 5 | Gemini 3 Flash (Preview) | 75 | #90 |
| 6 | Gemma 4 26B A4B | 75 | #360 |
| 7 | GLM-5 | 72.92 | #149 |
| 8 | Gemma 4 31B | 66.67 | #212 |
| 9 | GPT-5.4 | 62.5 | #83 |
| 10 | Claude Opus 4.6 | 62.5 | #67 |
| 11 | Gemini 2.5 Flash | 60.42 | #255 |
| 12 | Claude Opus 4.5 | 58.33 | #87 |
| 13 | Gemini 3.1 Flash Lite (Preview) | 58.33 | #201 |
| 14 | Gemini 2.5 Pro | 52.08 | #159 |
| 15 | Claude Sonnet 4.5 | 52.08 | #148 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kaggle-time-traveller-s-dilemma · How It Works · Data refreshed daily, snapshot 2026-11-05.