MultiWikiQHalluA - Danish: leaderboard

Faithfulness hallucination in Danish reading-comprehension answers (MultiWikiQHalluA, 2026): the share of 1,000 MultiWikiQA test answers in which a fine-tuned mmBERT-small LettuceDetect token classifier flags at least one token as unsupported by the Wikipedia context. Percent of answers; lower is better.

Metric: Answer-level hallucination rate (%): share of the 1,000 Danish MultiWikiQA test answers in which the fine-tuned mmBERT-small LettuceDetect token classifier flags at least one token as unsupported by the Wikipedia context, times 100; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1cogito-v1-preview-llama-70B8#927
2cogito-v1-preview-qwen-32B9#1181
3Qwen 3 14B12#1269
4Gemma 3 12B (IT)13#1887
5Qwen 3 0.6B52#8350

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=multiwikiqhallua-danish · How It Works · Data refreshed daily, snapshot 2026-11-21.