Coding benchmark score

Vals MedCode: 2,755 ICD-10-CM samples; Claude Opus 5 63.57% (July mirror).

Who judgesAutomatable
EvidenceGrade E2
Product typeCoding and RCMID R2

What coding benchmark score means

Vals MedCode: 2,755 ICD-10-CM samples; Claude Opus 5 63.57% (July mirror).

Why it matters. Public benchmarks give a rough comparison, though they differ from your notes.

Example, illustrative

A benchmark of 2,755 ICD-10-CM samples scores each model.

How to measure it

01

Create ground truth

Certified coders code a sample of notes.

02

Compare at two levels

Report accuracy per code and per claim.

03

Separate over and under

Overcoding and undercoding carry different risks.

04

Apply payer rules

Run deterministic checks on code validity and policy.

Watch out for

Vendors often quote one accuracy figure without saying how it was measured.

What a good report shows

Accuracy against certified-coder audit, by code set, with over and under coding shown apart.

Report accuracy against certified-coder audit, by code set, with overcoding and undercoding shown separately and the payer rules applied.

Reviewed by the Precision research team.

How strong is the evidence

E2 · Preprint, benchmark page or vendor-run result. Backed by numbers from a preprint, a benchmark page or a vendor-run test. Treat as strong but not settled.

What the sources say

(Vals).

  • Automatable. Can be scored by software, including an AI grader that is spot-checked against clinicians.
  1. Valsvals.ai

Found something out of date? Write to hello@precision.example.

Quick answers

What is coding benchmark score in healthcare AI?

Vals MedCode: 2,755 ICD-10-CM samples; Claude Opus 5 63.57% (July mirror). It sits in the coding, prior authorisation and revenue cycle group.

Why does coding benchmark score matter?

Public benchmarks give a rough comparison, though they differ from your notes. A benchmark of 2,755 ICD-10-CM samples scores each model.

How is coding benchmark score measured?

Create ground truth, then Compare at two levels, then Separate over and under, then Apply payer rules. Report accuracy against certified-coder audit, by code set, with overcoding and undercoding shown separately and the payer rules applied.

What should I watch out for with coding benchmark score?

Vendors often quote one accuracy figure without saying how it was measured. A good report gives: accuracy against certified-coder audit, by code set, with over and under coding shown apart.

Can coding benchmark score be automated?

Largely yes, including AI graders that are spot-checked against clinicians. Report the grader's error rate beside every score.

Measure coding benchmark score on your own data