TurkBench - Overall: leaderboard
TurkBench (METU NLP group and Turkcell, SIGTURK 2026): mean of 20 Turkish task scores spanning knowledge, language understanding, reasoning, content moderation, grammar and vocabulary, and instruction following. Scale 0-100, higher is better; only models scored on every task are included, since the site averages whatever a model has.
Metric: Mean task score (0-100). Source: huggingface.co. Saturation forecast: Around February 2027. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 397B A17B | 85.74 | #154 |
| 2 | Qwen 3.5 122B A10B | 85.35 | #184 |
| 3 | Qwen 3.5 35B A3B | 84.09 | #268 |
| 4 | GLM-5 FP8 | 83.11 | #147 |
| 5 | Qwen 3.5 27B | 80.52 | #192 |
| 6 | MiniMax-M2.5 | 80.51 | #316 |
| 7 | MiniMax-M2 | 79.74 | #327 |
| 8 | GPT-OSS-120B | 78.63 | #346 |
| 9 | DeepSeek V3.2 Exp | 77.14 | #241 |
| 10 | GLM-4.6 | 76.88 | #265 |
| 11 | Step 3.5 Flash | 75.65 | #281 |
| 12 | DeepSeek V3.1 | 75.19 | #275 |
| 13 | Qwen 3 Next 80B A3B Instruct | 74.95 | #381 |
| 14 | GLM-4.7 Flash | 74.95 | #543 |
| 15 | Qwen 3 30B A3B 2507 Instruct | 73.41 | #492 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=turkbench-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.