DoRA (Defense Documents) - Hallucination (GTE Retrieval): leaderboard

DoRA (2026) builds a retrieval-augmented question answering benchmark from defense-related documents, with test questions in five intent styles and auditable evidence. RAGEval keypoint hallucination rate in percent with GTE-retrieved passages as context, judged by an LLM; lower is better.

Metric: RAGEval keypoint hallucination rate (%) of grounded answers to the 1,259-question DoRA defense-document test split with the top-3 passages retrieved by GTE (gte-multilingual-base) as context: per-question share of the reference answer's keypoints the answer addresses incorrectly, judged by GPT-4o-mini and averaged over questions; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-3.5 Turbo3.71#3005
2Llama 3.1 8B Instruct4.44#4600
3Qwen 2.5 7B Instruct4.65#2723
4Ministral-8B-Instruct-24104.7#4555
5GPT-4o Mini4.92#1667
6Llama 3 8B Instruct5.24#4852
7GPT-4o5.6#751
8Claude 3.5 Haiku5.87#1580

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

Interactive version: theaggregate.ai/benchmark?slug=dora-defense-documents-hallucination-gte-retrieval · How It Works · Data refreshed daily, snapshot 2026-11-21.