KGHaluBench - Breadth Hallucination Rate: leaderboard
Metric: Breadth-of-knowledge hallucination rate (%): share of non-abstained responses that the entity-level filter classifies as describing the wrong entity, over 10 assessments of 150 questions each (compound questions about Wikidata entities, generated dynamically from the knowledge graph), default sampling parameters; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 7.69 | #176 |
| 2 | GPT-5 | 8.2 | #91 |
| 3 | Grok 4 | 10.27 | #169 |
| 4 | Claude Opus 4.1 (20250805) | 10.32 | #126 |
| 5 | GPT-4.1 | 10.84 | #240 |
| 6 | Llama 3.1 405B Instruct | 10.88 | #447 |
| 7 | Claude Sonnet 4 (20250514) | 13.01 | #211 |
| 8 | Gemini 2.5 Pro | 14.18 | #145 |
| 9 | Grok 3 | 15.2 | #296 |
| 10 | GLM-4.5 | 16.24 | #265 |
| 11 | Qwen 3 235B A22B 2507 Instruct | 21.49 | #291 |
| 12 | Gemini 2.5 Flash | 21.55 | #237 |
| 13 | GPT-4.1 Mini | 21.62 | #346 |
| 14 | GPT-OSS-120B | 22.74 | #330 |
| 15 | Kimi K2 | 23.66 | #236 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=kghalubench-breadth-hallucination-rate · How It Works · Data refreshed daily, snapshot 2026-10-11.