Grip on LLMs - HonestCityBench - Narrow Expertise: leaderboard
Grip on LLMs (arXiv 2608.09925v1, Appendix Table 2): HonestCityBench, 530 Dutch prompts that a model cannot or should not answer directly. Share of responses judged to appropriately acknowledge that the prompt needs narrow domain expertise, percent; most models score near zero. Higher is better.
Metric: Appropriate acknowledgement rate (%; LLM judge). Source: arxiv.org. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 3 12B (IT) | 12 | #672 |
| 2 | Llama 3.1 8B Instruct | 8 | #1017 |
| 3 | Gemma 3 27B (IT) | 5 | #528 |
| 4 | GEITje-7B-ultra | 5 | #1435 |
| 5 | GPT-4o | 4 | #357 |
| 6 | GPT-5 Mini | 4 | #193 |
| 7 | OLMo-2-0325-32B-Instruct | 4 | #847 |
| 8 | EuroLLM-9B-Instruct | 3 | #1047 |
| 9 | SmolLM3-3B | 3 | #1239 |
| 10 | GPT-5 | 1 | #105 |
| 11 | GPT-4o Mini | 1 | #598 |
| 12 | Qwen 3 8B | 1 | #678 |
| 13 | Qwen 3 32B | 1 | #449 |
| 14 | GPT-OSS-20B | 1 | #523 |
| 15 | Mistral 7B Instruct (v0.3) | 1 | #1255 |
Interactive version: theaggregate.ai/benchmark?slug=grip-on-llms-honestcitybench-narrow-expertise · How It Works · Data refreshed daily, snapshot 2026-11-05.