Start building

Benchmarks

Ranked #1 across the benchmarks that matter

Independent benchmarks across medical coding, speech recognition, clinical reasoning, and agentic AI. Tested head-to-head against the largest AI labs in the world.

Get API keyMeet with an Expert

The clinical AI stack

Significant outperformance across the full clinical-reasoning stack

Corti Symphony leads the next best competitor across agent reasoning, medical coding, and speech accuracy. The hardest problems in healthcare won’t be solved by the biggest models. They’ll be solved by specialized, clinical-grade models, validated in production.

What the comparison measures

Agent reasoning
HealthBench Professional
Medical coding
Automated coding F1
Speech accuracy
Clinical word accuracy

Corti Symphony vs the next-best system

Agent reasoning, medical coding and speech accuracy

How to read the evidence

Run on real clinical tasks, against real inputs, head-to-head with the largest AI labs in the world.

  • Stress tests, not averages.

    The dataset deliberately overrepresents difficult and adversarial examples by roughly 3.5x. A 60% score here can coexist with strong performance in typical clinical use.

  • Use the right measure.

    Higher is better for accuracy and F1. Lower is better for word error rate (WER), where accuracy = 100% − WER.

  • Published so you can verify.

    The results are published so you can verify them yourself.

01 — Agents

Outperforming every major LLM on OpenAI’s clinical benchmark

The highest performance across all major models on HealthBench Professional.

#1 on HealthBench Professional

HealthBench Professional is an open benchmark developed by OpenAI to evaluate LLMs on the kinds of tasks clinicians actually bring to AI in practice, scored against rubrics written and adjudicated by three or more physicians. Corti Symphony scores 60.5% overall, above all other models including OpenAI’s GPT-5.4, Claude’s Opus 4.7, and ChatGPT for Clinicians, OpenAI’s own direct-to-clinician product, which scores 59% on the same benchmark.

60.5%
Corti Symphony overall score
59%
ChatGPT for Clinicians, OpenAI’s clinician product
3+
Physicians adjudicate each rubric

HealthBench Professional overall score comparison

Physician-adjudicated benchmark across real clinical interactions.

The safest model under the most difficult conditions

Red teaming is the adversarial subset of HealthBench Professional, where clinicians deliberately attempt to surface failure modes rather than test routine usage. Roughly one third of the benchmark consists of red teaming examples. Corti Symphony scores 59% on red teaming overall, nearly double GPT-5.4’s 30.3%, with a perfect 100% in the research category against GPT-5.4’s 64%.

59.0%
Symphony red-teaming score
30.3%
GPT-5.4 red-teaming score
100%
Symphony score on research prompts

HealthBench Professional red teaming score comparison

Adversarial evaluation across care consult, research, and writing tasks.

02 — Medical coding

Leading automated medical coding benchmarks

Ranked #1 on performance across synthetic, academic, and real-world clinical data.

Accuracy comparison

F-score is the harmonic mean of precision and recall, used to evaluate medical coding accuracy across diagnostic and procedural codes. Corti leads with a score of 74%, outperforming Anthropic (59%), Amazon (58%), OpenAI (48%), and Google (43%).

74%
F1 accuracy
>25%
Improvement vs. the next-best system
#1
On noisy clinical datasets

Medical coding accuracy

Automated coding F1, higher is better

03 — Speech to text

The most accurate medical speech to text API

Validated on 150,000+ medical terms across 14 languages and every clinical specialty.

90%+ reduction in Word Error Rate compared to the next-best STT model

Measured on clinical audio across realtime and async configurations against generalist STT APIs and open-source models. Symphony reaches 98.6% word accuracy in English, a 1.4% word error rate against 16.4% for AWS (offline), the next-best.

98.6%
Symphony word accuracy
83.6%
Next-best word accuracy (AWS, offline)
150,000+
Medical terms validated

Word accuracy rate comparison

Realtime and async configurations in English.

Symphony outperforms every benchmarked model across English, German, and French

Measured on realtime configurations. Accuracy holds in every tested language.

98.6%
English word accuracy
97.6%
German word accuracy
96.1%
French word accuracy

Word accuracy rate comparison

Realtime configurations in English, German, and French.

20% fewer errors than Dragon Medical One

Symphony delivers better dictation accuracy at a fraction of the cost, with full data sovereignty for European healthcare organizations. It gives development teams a foundation to build clinical documentation tools without the lock-in.

20%
Fewer errors with Corti Symphony
4.1%
Corti dictation WER
5.7%
DMO dictation WER

Dictation word error rate

Lower is better

Leading on formatting accuracy across every tested language

Punctuation, capitalization, and structure, handled correctly without post-processing.

98.3%
English formatting accuracy
99.4%
German formatting accuracy
92.4%
French formatting accuracy

Formatting accuracy rate comparison

Realtime configurations in English, German, and French.

Keyterm biasing cuts missed medical terms in half

When Symphony is biased toward a vocabulary of expected medical terms, it catches significantly more of them without introducing false positives. Across both tested datasets, the false-negative rate for medical terms drops by roughly half while precision holds steady above 96%. The vocabulary travels with the request, requires no retraining, and works across any supported language or configuration.

47.9%
Fewer missed medical terms (MedDictate, EN)
50.9%
Fewer missed medical terms (MedRad, EN)
96%+
Precision with biasing

Precision with keyterm biasing

Precision with and without biasing, higher is better

Symphony can detect when audio quality decreases

Symphony continuously monitors audio quality during transcription and flags segments where signal conditions are likely to degrade accuracy. As noise increases, word error rate rises sharply and flagging tracks it, giving developers a reliable signal to trigger fallback logic, prompt re-recording, or surface quality warnings to clinicians before a bad transcript reaches a downstream workflow.

0%
Segments flagged in clean audio
40%
Segments flagged at 6 dB SNR
100%
Segments flagged at 0 dB SNR

Segment WER and % segments flagged vs SNR

Audio quality event detection under increasing noise conditions.

04 — Clinical reasoning

The best fact extraction tool for ambient AI in healthcare

FactsR™ extracts, refines, and validates clinical facts as the consultation unfolds.

FactsR™ Groundedness Accuracy

Groundedness measures whether the content in a generated note can be traced back to what was actually said in the consultation. FactsR™ maintains groundedness at 94%, comfortably above the level seen in clinician-written reference notes on the Primock57 benchmark.

94%
FactsR™ groundedness

Groundedness accuracy

Higher is better

FactsR™ Conciseness Errors

Conciseness measures how much extraneous detail ends up in a clinical note. FactsR™ reduces conciseness errors by 86%, bringing them down from 14.3% to 2.0%, by validating each extracted fact before it reaches the note.

86%
Reduction in conciseness errors
2.0%
Conciseness errors with FactsR™

Conciseness errors

Lower is better

FactsR™ Missed Findings

Completeness measures whether clinically relevant information from the consultation makes it into the final note. FactsR™ reduces completeness errors by 49%, with missing content dropping from 23.3% to 11.7%, through its real-time extraction and refinement loop.

49%
Reduction in missed findings
11.7%
Missed findings with FactsR™

Missed findings

Lower is better

The clinical stack

Production-grade building blocks built for every layer of the clinical stack.

Ranked #1 on automated coding benchmarks, the lowest word error rate on clinical speech benchmarks, and the top HealthBench Professional score.

More from Corti on Benchmarks

View all

Frequently asked questions

Meet with an Expert

No. The dataset deliberately overrepresents difficult and adversarial examples by roughly 3.5x. A 60% score here can coexist with strong performance in typical clinical use. The benchmark is a stress test, not an average-case measurement.

About a third of the benchmark consists of physicians deliberately trying to break models. Strategies include false premises, role-play framing, and presenting questionable diagnoses as fact. Symphony scores 59.0 on this subset against GPT-5.4’s 30.3. That gap reflects robustness under adversarial clinical pressure, not just routine performance.

Three categories reflecting real clinical workflows: care consult (differential diagnosis, treatment reasoning), writing and documentation (note generation, coding, patient messaging), and medical research (synthesizing and finding clinical evidence).

Yes. Symphony is built for healthcare environments, including support for sovereign cloud deployments for organisations with strict data residency requirements.

2% Word Error Rate (WER) on clinical speech across 150,000+ medical terms and 14 languages. In practical terms, that’s three to four times fewer errors than the next best alternatives.

Word Error Rate (WER) measures the accuracy of automatic speech recognition (ASR) systems by calculating how often a model substitutes, inserts, or deletes words compared to a correct reference transcript. It’s expressed as a percentage, where lower is better and 0% is a perfect transcription. Mathematically: Accuracy = 100% - WER.

It runs in real time during the consultation rather than after. By the time the visit ends, structured facts are already extracted, validated, and timestamped.