MedCode: leaderboard
ICD-10-CM diagnosis coding from de-identified discharge summaries, progress and consult notes, scored on 2,755 primary and secondary codes; Vals AI with Harvard Medical School and Protege AI.
Metric: Accuracy (%). Source: www.vals.ai. Status: saturation imminent. 90 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5 (Max) | 63.57 |
| 2 | Gemini 3.1 Pro (Preview) (High) | 59.06 |
| 3 | Claude Fable 5 (Max) | 56.07 |
| 4 | Gemini 3 Flash (Preview) (High) | 55.92 |
| 5 | Gemini 3.5 Flash (High) | 55.83 |
| 6 | Claude Opus 4.7 (Max) | 54.86 |
| 7 | Claude Fable 5.1 (Max) | 53.51 |
| 8 | Gemini 3.7 Flash (High) | 53.39 |
| 9 | Claude Opus 4.8 (Max) | 53.22 |
| 10 | Gemini 3.6 Flash (High) | 53.15 |
| 11 | GPT-5.1 (High) | 52.73 |
| 12 | Gemini 3 Pro (Preview) (High) | 52.2 |
| 13 | Muse Spark | 51.31 |
| 14 | Gemini 2.5 Pro | 50.59 |
| 15 | GPT-5.2 (xHigh) | 49.75 |
Interactive version: theaggregate.ai/benchmark?slug=medcode · How It Works · Data refreshed daily, snapshot 2026-09-05.