CliniCARE-Bench - Defect-Free Accuracy: leaderboard
CliniCARE-Bench (arXiv 2608.07796v1, Table 3): agent systems audit de-identified EHR records against a clinical policy corpus across 25 clinician-built scenarios (750 cases, one run each) and return Yes, No, or Indeterminate (lack of data or medically ambiguous). Share of report-present runs with a correct verdict and no judged defect (GPT-5.5 judge), in percent; higher is better. Rows are harness plus model systems.
Metric: Report-present runs with correct verdict and no judged defect (%). Source: arxiv.org. Saturation forecast: Around September 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 71.3 | #33 |
| 2 | Claude Opus 5 | 70 | #16 |
| 3 | GPT-5.6 Sol | 67.5 | #22 |
| 4 | GPT-5.4 | 65.6 | #83 |
| 5 | Claude Sonnet 5 | 65.6 | #65 |
| 6 | GLM-5.2 | 64.8 | #80 |
| 7 | GPT-5.6 Luna | 60.8 | #79 |
| 8 | Gemini 3.5 Flash | 60 | #66 |
| 9 | Gemini 3.6 Flash | 59.1 | #57 |
| 10 | DeepSeek V4 Pro | 58.7 | #112 |
| 11 | DeepSeek V4 Flash | 58.3 | #155 |
| 12 | Gemini 3.1 Pro (Preview) | 57.9 | #64 |
| 13 | GPT-5.4 Mini | 56.1 | #228 |
| 14 | MiniMax-M3 | 54.7 | #151 |
| 15 | Kimi K2.7 Code | 54.1 | #95 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=clinicare-bench-defect-free-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-05.