CliniCARE-Bench - Verdict Macro-F1: 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). Macro-F1 across the four verdict classes, 0-1; higher is better. Rows are harness plus model systems.
Metric: Macro-F1 over the four verdict classes (0-1). Source: arxiv.org. Saturation forecast: Around September 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 0.7 | #16 |
| 2 | GPT-5.5 | 0.7 | #33 |
| 3 | Gemini 3.1 Pro (Preview) | 0.66 | #64 |
| 4 | GPT-5.6 Sol | 0.66 | #22 |
| 5 | Claude Sonnet 5 | 0.65 | #65 |
| 6 | Gemini 3.6 Flash | 0.65 | #57 |
| 7 | GLM-5.2 | 0.64 | #80 |
| 8 | GPT-5.4 | 0.63 | #83 |
| 9 | Gemini 3.5 Flash | 0.62 | #66 |
| 10 | DeepSeek V4 Pro | 0.61 | #112 |
| 11 | GPT-5.6 Luna | 0.59 | #79 |
| 12 | DeepSeek V4 Flash | 0.59 | #155 |
| 13 | Qwen 3.7 Plus | 0.58 | #98 |
| 14 | GPT-5.4 Mini | 0.56 | #228 |
| 15 | Kimi K2.7 Code | 0.56 | #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-verdict-macro-f1 · How It Works · Data refreshed daily, snapshot 2026-11-05.