Grip on LLMs - Dutch tinyARC: leaderboard
Grip on LLMs (From Values to Benchmarks, arXiv 2608.09925v1, Appendix D), a Dutch governmental evaluation suite built with the City of Amsterdam. GP-IRT accuracy estimate from the 100 tinyARC-Challenge science reasoning items in a Dutch machine translation, percent. Higher is better.
Metric: Accuracy (%; GP-IRT estimate from 100 tinyBenchmarks items, Dutch machine translation). Source: arxiv.org. Saturation forecast: Around February 2027. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 84 | #105 |
| 2 | Qwen 3 32B | 84 | #449 |
| 3 | Mistral Medium 3 | 82 | #507 |
| 4 | Gemma 3 27B (IT) | 81 | #528 |
| 5 | GPT-4o | 80 | #357 |
| 6 | Mistral Large 3 | 80 | #410 |
| 7 | Mistral Small 3 | 79 | #630 |
| 8 | GPT-4o Mini | 78 | #598 |
| 9 | DeepSeek R1 Distill Qwen 32B | 75 | #662 |
| 10 | aya-expanse-32B | 72 | #687 |
| 11 | OLMo-2-0325-32B-Instruct | 70 | #847 |
| 12 | Gemma 3 12B (IT) | 69 | #672 |
| 13 | Qwen 3 8B | 68 | #678 |
| 14 | GPT-5 Mini | 66 | #193 |
| 15 | GPT-5 Nano | 64 | #439 |
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-tinyarc · How It Works · Data refreshed daily, snapshot 2026-11-05.