Clinical Verbatim Citation - Verbatim Compliance: leaderboard
Share of a model's citations in clinical QA that quote the reference material verbatim (exact, normalised or elided), so a clinician can check them in seconds. Higher is better. Source: arxiv.org/abs/2609.15964.
Metric: Verbatim-compliant citations (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 100 | #16 |
| 2 | DeepSeek V4 Pro | 99.2 | #112 |
| 3 | GPT-5 Mini | 99.1 | #193 |
| 4 | Kimi K3 | 99.1 | #35 |
| 5 | Claude Sonnet 5 | 98.9 | #65 |
| 6 | DeepSeek V4 Flash | 98.8 | #155 |
| 7 | GLM-5.2 | 97.6 | #80 |
| 8 | GPT-5.4 | 97.4 | #83 |
| 9 | GPT-5.4 Mini | 94.5 | #228 |
| 10 | GLM-4.7 Flash | 78.8 | #543 |
| 11 | Claude Haiku 4.5 | 69 | #287 |
| 12 | GPT-4.1 Mini | 66.2 | #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-verbatim-compliance · How It Works · Data refreshed daily, snapshot 2026-11-05.