Grip on LLMs - Dutch Simplification Average: leaderboard
Grip on LLMs (From Values to Benchmarks, arXiv 2608.09925v1, Appendix D), a Dutch governmental evaluation suite built with the City of Amsterdam. Dutch text simplification: mean SARI over City of Amsterdam communication sentence pairs and a filtered INT Duidelijke Taal subset, 0-100. Higher is better.
Metric: SARI (0-100; mean of two datasets). Source: arxiv.org. Saturation forecast: Not forecast. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GEITje-7B-ultra | 45.77 | #1435 |
| 2 | OLMo-2-0325-32B-Instruct | 43.71 | #847 |
| 3 | Mistral Large 3 | 43.7 | #410 |
| 4 | Gemma 3 27B (IT) | 43.1 | #528 |
| 5 | OLMo-2-1124-7B-Instruct | 42.38 | #1343 |
| 6 | aya-expanse-32B | 41.56 | #687 |
| 7 | Phi-4 Mini Instruct | 40.99 | #1232 |
| 8 | GPT-4o | 40.91 | #357 |
| 9 | EuroLLM-9B-Instruct | 40.78 | #1047 |
| 10 | GPT-5 | 40.74 | #105 |
| 11 | Gemma 3 12B (IT) | 40.6 | #672 |
| 12 | Llama 3.1 8B Instruct | 40.44 | #1017 |
| 13 | Mistral 7B Instruct (v0.3) | 39.7 | #1255 |
| 14 | Qwen 3 32B | 39.54 | #449 |
| 15 | Mistral Medium 3 | 39.48 | #507 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=grip-on-llms-dutch-simplification-average · How It Works · Data refreshed daily, snapshot 2026-11-05.