EuroEval German Summarization - Mlsum DE: leaderboard
EuroEval's German summarization task on the Mlsum DE dataset — models write abstractive summaries of German news articles, scored with automatic text-overlap metrics, from the EuroEval European language model benchmark suite.
Metric: chrF3++ (%). Source: euroeval.com. Saturation forecast: Not forecast. 149 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gemma-3-27B-pt | 41.19 | #502 |
| 2 | Llama 3.1 70B | 39.19 | #592 |
| 3 | Llama 3.1 8B | 38.51 | #1151 |
| 4 | gemma-3-12B-pt | 38.23 | #686 |
| 5 | occiglot-7B-eu5-instruct | 37.96 | |
| 6 | occiglot-7B-eu5 | 37.82 | |
| 7 | EuroLLM-22B-Instruct-2512 | 37.54 | #701 |
| 8 | SmolLM3-3B-Base | 37.45 | #1285 |
| 9 | Llama 3.3 70B Instruct | 36.95 | #534 |
| 10 | Mistral Small 3.1 | 36.83 | #610 |
| 11 | Voxtral-Small-24B-2507 | 36.76 | #581 |
| 12 | gemma-3-4B-pt Base | 36.49 | #1134 |
| 13 | Llama 3.1 8B Instruct | 36.39 | #1035 |
| 14 | Granite 4.0 Micro | 36.08 | #882 |
| 15 | Mistral Small 3.2 | 36.03 | #605 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=euroeval-german-summarization-mlsum-de · How It Works · Data refreshed daily, snapshot 2026-10-20.