TurkBench - Summarization: leaderboard
TurkBench (METU NLP group and Turkcell, SIGTURK 2026) task: 262 Turkish texts to summarize concisely without losing the main idea, graded by an LLM judge (0-100, higher is better).
Metric: LLM-as-a-judge score (0-100). Source: huggingface.co. Saturation forecast: Around February 2027. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MiniMax-M2.5 | 84.09 | #316 |
| 2 | MiniMax-M2 | 83.16 | #327 |
| 3 | Qwen 3 30B A3B 2507 Instruct | 81.87 | #492 |
| 4 | GPT-OSS-120B | 81.82 | #346 |
| 5 | DeepSeek V3.2 Exp | 81.82 | #241 |
| 6 | Qwen 3 Next 80B A3B Instruct | 81.7 | #381 |
| 7 | Step 3.5 Flash | 80.72 | #281 |
| 8 | DeepSeek V3.1 | 80.13 | #275 |
| 9 | Gemma 3 12B (IT) | 77.47 | #672 |
| 10 | aya-expanse-8B | 76.97 | #914 |
| 11 | GLM-4.6 | 76.45 | #265 |
| 12 | Qwen 3.5 27B | 76.41 | #192 |
| 13 | Qwen 3.5 35B A3B | 75.59 | #268 |
| 14 | Qwen 2.5 14B Instruct | 75 | #654 |
| 15 | Qwen 3.5 122B A10B | 74.5 | #184 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=turkbench-summarization · How It Works · Data refreshed daily, snapshot 2026-11-05.