HalluScore - Adversarial: leaderboard
Metric: Factual hallucination rate (%) on the 360 adversarial HalluScore questions (misleading or carefully crafted phrasing), zero-shot short-answer Arabic question answering at temperature 0 with no retrieval, responses annotated by humans; an answer that states an unsupported fact with confidence counts, an expressed uncertainty or refusal does not; lower is better. Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 31.11 |
| 2 | Claude Opus 4 | 34.72 |
| 3 | Claude Sonnet 4.5 | 38.89 |
| 4 | GPT-4o | 44.72 |
| 5 | O4 Mini | 51.11 |
| 6 | Grok 4 Fast (Reasoning) | 55.56 |
| 7 | Qwen 3 Next 80B A3B Instruct | 58.61 |
| 8 | DeepSeek R1 | 60.83 |
| 9 | DeepSeek V3 | 62.5 |
| 10 | Grok 4 | 65.28 |
| 11 | Llama 4 Maverick Instruct FP8 | 70.56 |
| 12 | ALLaM-7B-Instruct-preview | 75 |
| 13 | Fanar-1-9B | 80.83 |
| 14 | O3 | 81.67 |
Interactive version: theaggregate.ai/benchmark?slug=halluscore-adversarial · How It Works · Data refreshed daily, snapshot 2026-10-07.