Clinical Verbatim Citation - Citation Coverage: leaderboard
Clinical question answering that must be verifiable by construction: the share of claim sentences that carry a citation (attribution window k=1). Higher is better. Source: arxiv.org/abs/2609.15964.
Metric: Claims with a citation (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 100 | #193 |
| 2 | GPT-5.4 Mini | 100 | #228 |
| 3 | GPT-5.4 | 99.8 | #83 |
| 4 | DeepSeek V4 Pro | 98.5 | #112 |
| 5 | DeepSeek V4 Flash | 98.5 | #155 |
| 6 | Kimi K3 | 98.5 | #35 |
| 7 | Claude Opus 5 | 98 | #16 |
| 8 | Claude Sonnet 5 | 95 | #65 |
| 9 | GLM-5.2 | 92.1 | #80 |
| 10 | Claude Haiku 4.5 | 74.8 | #287 |
| 11 | GLM-4.7 Flash | 66.1 | #543 |
| 12 | GPT-4.1 Mini | 49.6 | #363 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clinical-verbatim-citation-citation-coverage · How It Works · Data refreshed daily, snapshot 2026-11-05.