Deep-Research Citation Benchmark - Factual Support: leaderboard
Metric: Cohen kappa (-1 to 1) between the LLM judge and human-reviewed gold labels on the factual support rubric criterion of 624 attribution-citation pairs parsed from an adversarial 25-domain deep-research report (scores binarized at 0.5); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 0.7 |
| 2 | Claude Sonnet 4.6 | 0.68 |
| 3 | Gemini 3.1 Flash Lite (Preview) | 0.65 |
| 4 | GPT-5 Mini | 0.65 |
| 5 | Claude Haiku 4.5 | 0.63 |
| 6 | Gemini 3.1 Pro (Preview) | 0.62 |
| 7 | GPT-OSS-120B | 0.59 |
| 8 | GPT-5.4 Mini | 0.58 |
Interactive version: theaggregate.ai/benchmark?slug=deep-research-citation-benchmark-factual-support · How It Works · Data refreshed daily, snapshot 2026-09-29.