AIME-TR: leaderboard
AIME 2024 and 2025 translated into Turkish — the same 60 competition problems, asked in a language the training corpus has far less of. Read against the English AIME boards it measures how much of a model's competition maths survives translation; the roster is small and open-weight-heavy.
Metric: Accuracy (%, AIME 2024+2025 in Turkish). Source: huggingface.co. Saturation forecast: Around January 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 27B | 76.67 | #187 |
| 2 | Qwen 3.5 35B A3B | 71.67 | #256 |
| 3 | Qwen 3.5 9B | 63.33 | #379 |
| 4 | Qwen 3.5 4B | 53.33 | #475 |
| 5 | Gemma 3 27B (IT) | 16.67 | #523 |
| 6 | Gemma 3 12B (IT) | 13.33 | #664 |
| 7 | Gemma 3 4B (IT) | 3.33 | #992 |
| 8 | Qwen 3.5 2B | 3.33 | #1148 |
| 9 | Gemma 3 1B (IT) | 0 | #1504 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=aime-tr · How It Works · Data refreshed daily, snapshot 2026-10-20.