TurkBench - Topic Detection: leaderboard
TurkBench (METU NLP group and Turkcell, SIGTURK 2026) task: 240 multiple-choice items identifying the main topic of Turkish sociology texts. Accuracy 0-100, higher is better.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 35B A3B | 89.71 | #268 |
| 2 | Gemma 3 27B (IT) | 88.07 | #528 |
| 3 | Qwen 3.5 397B A17B | 87.24 | #154 |
| 4 | Qwen 3.5 27B | 86.42 | #192 |
| 5 | Qwen 3.5 122B A10B | 86.01 | #184 |
| 6 | Qwen 3 Next 80B A3B Instruct | 85.6 | #381 |
| 7 | MiniMax-M2.5 | 85.6 | #316 |
| 8 | DeepSeek V3.2 Exp | 85.6 | #241 |
| 9 | MiMo-V2-Flash | 85.6 | #378 |
| 10 | GPT-OSS-120B | 85.59 | #346 |
| 11 | GLM-5 FP8 | 85.19 | #147 |
| 12 | Qwen 3 32B | 85.18 | #449 |
| 13 | Qwen 3 30B A3B 2507 Instruct | 85.18 | #492 |
| 14 | Gemma 2 9B (IT) | 84.77 | #795 |
| 15 | DeepSeek V3.1 | 84.77 | #275 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=turkbench-topic-detection · How It Works · Data refreshed daily, snapshot 2026-11-05.