Benchmarks › Vals MedCode

Vals MedCode

Assign medical diagnosis codes from patient records
63.6top score · Claude Opus 5
78models ranked
28open-weights models
Diagnosis-code accuracyhigher is better
Oct 20, 2026last updated

How the test works

1

Task

2,755 primary/secondary diagnosis-code labels

  • What the model doesAssign medical diagnosis codes from patient records
  • Kinds of tasksICD-10-CM medical coding
2

Run

30,000 generated tokens maximum

  • Temperature1
3

Score

Diagnosis-code accuracy

  • Graded byPredicted diagnosis codes compared with reconciled annotations from two certified coders
  • Better isHigher
Sources: vals.ai

Vals MedCode leaderboardtop 10 of 78

Full leaderboard with filters →
ModelVals MedCodePer taskPriceSpeedSupportsContext
Claude Opus 5Anthropic
63.6unstated
—$5 / $25—1M
59.1unstated
—$2 / $12—1M
58.8unstated
—$2 / $10—1M
56.1unstated
—$10 / $50—1M
55.9unstated
—$0.50 / $3—1M
54.9unstated
—$5 / $25—1M
53.5unstated
—$10 / $50—1M
53.2unstated
—$5 / $25—1M
53.2unstated
—$0.75 / $3.75—1M
52.9unstated
—$2 / $10—1M

Vals MedCode questions

What is Vals MedCode?›

Assign medical diagnosis codes from patient records. It has 2,755 primary/secondary diagnosis-code labels; patient-record count not stated. Created by Vals AI and Protege.

Which model is best on Vals MedCode?›

Claude Opus 5 leads Vals MedCode with 63.6, ahead of Gemini 3.1 Pro Preview (59.1) and Gemini 4 Argon (58.8).

What is the best open-weights model on Vals MedCode?›

Kimi K3 with 49.4, then MiniMax M3 (46.3).

How is Vals MedCode scored?›

Diagnosis-code accuracy; predicted diagnosis codes compared with reconciled annotations from two certified coders.