TurkBench - Parts of Speech: leaderboard
TurkBench (METU NLP group and Turkcell, SIGTURK 2026) task: 260 multiple-choice part-of-speech tagging items on Turkish news sentences, run with thinking off. 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 | DeepSeek V3.2 Exp | 83.77 | #241 |
| 2 | DeepSeek V3.1 | 83.6 | #275 |
| 3 | Gemma 3 27B (IT) | 81.81 | #528 |
| 4 | Qwen 3 Next 80B A3B Instruct | 81.3 | #381 |
| 5 | GLM-5 FP8 | 79.71 | #147 |
| 6 | Qwen 3.5 122B A10B | 79.42 | #184 |
| 7 | GLM-4.6 | 79.02 | #265 |
| 8 | Qwen 3.5 397B A17B | 78.81 | #154 |
| 9 | MiMo-V2-Flash | 78.39 | #378 |
| 10 | GPT-OSS-120B | 78.16 | #346 |
| 11 | Qwen 3.5 35B A3B | 78.1 | #268 |
| 12 | Qwen 3 30B A3B 2507 Instruct | 78.07 | #492 |
| 13 | Qwen 3 32B | 77.66 | #449 |
| 14 | MiniMax-M2 | 77.07 | #327 |
| 15 | Gemma 3 12B (IT) | 73.57 | #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-parts-of-speech · How It Works · Data refreshed daily, snapshot 2026-11-05.