Accuracy you can audit.
Measured the way medicine measures everything: blinded, against the human standard of care.
>90%
fewer errors than certified medical interpreters
Held to medicine's standard of proof.
- Blinded, head-to-head against certified interpreters
- Run by physician-researchers on real clinical dialogue
- Every omission, substitution, and distortion counted
Far fewer errors, in every language.
Total interpretation errors in a blinded evaluation: Opalite against three certified medical interpreters per language.
Cantonese
p = .01895–97% fewer
total interpretation errors
Mandarin
p = .00595–96% fewer
total interpretation errors
Spanish
p = .01198–99% fewer
total interpretation errors
Errors per evaluation
Cantonese
p = .018
Mandarin
p = .005
Spanish
p = .011
Blinded clinical evaluation of total interpretation error counts, Opalite vs. three certified medical interpreters per language.
Weighted for severity, the gap only widens.
Each error scored for clinical severity: minor counts 1, major 5, critical 10.
>98%
fewer major and critical errors than certified interpreters
Every major and critical error was eliminated in Spanish and Cantonese.
Severity-weighted score
- Cantonese
- 4vs 207-431p = .038
- Mandarin
- 10vs 490-636p = .007
- Spanish
- 1vs 206-279p = .007
Where certified interpretation broke down.
Among certified interpreters, the errors clustered in the two categories that matter most in a clinical visit.
33-44%
were information omissions
16-20%
were medical terminology errors
Opalite produced too few errors to categorize, with no single dominant error type.
Rated higher by blinded clinicians, chosen first every time.
A seven-domain blinded quality survey, scored on a 10-point scale.
9.0 / 10
quality rating in Cantonese
9.0 / 10
quality rating in Mandarin
9.0 / 10
quality rating in Spanish
Ranked first by the blinded annotators in all three languages.
Want the full study?
We'll walk your clinical team through it: method, error taxonomy, results.