Kaggle ECLeKTic: leaderboard
Google's ECLeKTic closed-book question set for cross-lingual knowledge transfer in 12 languages, built from facts that appear in one language's Wikipedia but not the others, on its Kaggle leaderboard. Overall success, the benchmark's main metric, credits a question only when answered correctly in both the source and a target language (0-100, higher is better).
Metric: Overall Success (%). Source: www.kaggle.com. Saturation forecast: Estimated already saturated. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 | 43.11 | #132 |
| 2 | Gemini 2.5 Pro (Preview 05-06) | 40.08 | #156 |
| 3 | O1 | 38.73 | #190 |
| 4 | Gemini 2.5 Pro (Preview 06-05) | 36.86 | #118 |
| 5 | GPT-4.1 | 34.16 | #258 |
| 6 | Gemini 2.5 Flash (Preview 05-20) | 30.85 | #229 |
| 7 | GPT-4o (2024-08-06) | 29.85 | #343 |
| 8 | Grok 3 | 28.98 | #313 |
| 9 | Gemini 2.0 Flash | 24.64 | #351 |
| 10 | Claude Opus 4 | 24.01 | #170 |
| 11 | Claude 3.7 Sonnet | 23.25 | #259 |
| 12 | O4 Mini | 22.59 | #185 |
| 13 | DeepSeek V3 | 21.99 | #329 |
| 14 | Gemini 1.5 Pro (002) | 20.24 | #368 |
| 15 | Claude 3.5 Sonnet (20241022) | 19.98 | #302 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kaggle-eclektic · How It Works · Data refreshed daily, snapshot 2026-11-05.