RadMatch - Radiologist Agreement: leaderboard
Absolute Kendall tau-b agreement with radiologist total-error counts on the pooled 200-pair ReXVal set, using the RadMatch pipeline. Count/enumerate baselines and RadEvalExpert are separate evaluations.
Metric: Absolute Kendall tau-b vs radiologists (ReXVal). Source: arxiv.org. Saturation forecast: Around January 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 | 0.79 | #37 |
| 2 | GPT-4.1 | 0.76 | #248 |
| 3 | Kimi K2.6 | 0.75 | #104 |
| 4 | Gemma 4 31B | 0.75 | #200 |
| 5 | GPT-5.4 | 0.74 | #79 |
| 6 | GPT-5.2 | 0.74 | #110 |
| 7 | GPT-5.5 | 0.69 | #31 |
| 8 | Qwen 3.5 35B A3B | 0.69 | #256 |
| 9 | DeepSeek V4 Pro | 0.68 | #102 |
| 10 | GPT-5.4 Mini | 0.66 | #212 |
| 11 | MedGemma 1.5 4B | 0.47 | #789 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=radmatch-radiologist-agreement · How It Works · Data refreshed daily, snapshot 2026-10-20.