TurkBench - Turkish General Knowledge: leaderboard
TurkBench (METU NLP group and Turkcell, SIGTURK 2026) task: 200 expert-written multiple-choice questions on Turkish culture, history and daily life. Accuracy 0-100, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5 FP8 | 85.94 | #147 |
| 2 | Qwen 3.5 397B A17B | 84.9 | #154 |
| 3 | GLM-4.6 | 83.85 | #265 |
| 4 | Qwen 3.5 122B A10B | 81.25 | #184 |
| 5 | MiniMax-M2 | 80.21 | #327 |
| 6 | GPT-OSS-120B | 77.6 | #346 |
| 7 | DeepSeek V3.1 | 77.6 | #275 |
| 8 | MiniMax-M2.5 | 77.6 | #316 |
| 9 | DeepSeek V3.2 Exp | 77.6 | #241 |
| 10 | Qwen 3.5 27B | 75.52 | #192 |
| 11 | Qwen 3.5 35B A3B | 75.52 | #268 |
| 12 | Step 3.5 Flash | 75.52 | #281 |
| 13 | MiMo-V2-Flash | 75 | #378 |
| 14 | Gemma 3 27B (IT) | 74.65 | #528 |
| 15 | Gemma 3 12B (IT) | 71.35 | #672 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=turkbench-turkish-general-knowledge · How It Works · Data refreshed daily, snapshot 2026-11-05.