ICD-10 medical coding
A distilled clinical-coding specialist from the Pareta fleet.
Send a clinical note through the same model:"auto" endpoint you
use for everything else — Pareta routes it to the specialist, constrains the
output to valid codes, and verifies before serving.
106%
of GPT-5.5 accuracy on ICD-10 coding (micro-F1)
138×
lower cost than the same GPT-5.5 baseline
JSON out
codes returned as structured output, constrained at decode time — not parsed from prose
$0.05 / $0.10
per 1M input / output tokens — a typical chart codes for a fraction
of a cent · pricing
No integration beyond the one you already have
There is no separate coding API to learn. The same two-line change that
puts you on Pareta serves coding too:
completion = client.chat.completions.create(
model="auto",
messages=[{"role": "user", "content": chart_note + "\n\nCode this encounter (ICD-10-CM)."}],
)
Don't take the benchmark's word for it. Upload your own de-identified
charts in the dashboard and run the same evaluation we run — your data, our
specialist, frontier models side by side, scored identically.