Corti, maker of AI foundation models for healthcare, has released a new agentic model for medical coding that it says outperforms a number of Big Tech models.
Symphony for Medical Coding outperforms those from OpenAI, Anthropic, Amazon, Oracle and Google by over 25% in clinical accuracy benchmarks. The product, available via API, builds on Corti’s flagship model, Symphony, already used by 200 teams in the U.S. today. The company works with electronic health record vendors, virtual care platforms, practice management systems and life sciences organizations globally.
Medical coding is extremely complicated. The U.S. coding system, ICD-10, has about 70,000 diagnosis codes. The automated process converts clinical notes into structured data to inform reimbursement, research and policy.
Being able to capture the right codes amid nuance is crucial; errors can be costly both in terms of missed revenue and under-captured diagnoses. In a recent study of Danish patient data, Corti found three times as many suicide attempts as had been coded. Missed trends like these impact resource allocation and intervention design.
"Most AI systems fall short in medical coding because they treat it as labeling, not reasoning. Correct coding depends on evidence, context, hierarchy and guideline interpretation,” Lars Maaløe, Ph.D., chief technology officer and co-founder of Corti, said in an announcement. "We built Symphony for Medical Coding to follow the same decision process expert coders use, and that is why the performance gap is so meaningful.”
Symphony was evaluated on two widely used public benchmarks in medical coding research, according to Corti: ACI-BENCH and MDACE. These were created by independent academic teams and professional medical coders. Symphony for Medical Coding is also validated on real-world clinical data from a large U.S. health system spanning emergency and outpatient care settings. All models, including Corti’s and the Big Tech companies' to which it was compared, were evaluated under identical conditions and run five times each to ensure consistency and reproducibility.
Specifically, Corti's model outperformed Anthropic’s Claude Opus 4.6, OpenAI’s GPT-5.4 and Google’s Gemini 3.1 Pro. It was also compared against systems built on top of those foundation models, including Oracle Health & AI’s MedDCR (built on GPT) and AWS AI’s fine-tuned coding model (built on Claude).
Existing models that automate coding typically rely on memorizing patterns from annotated datasets through supervised or semi-supervised learning, per Corti. This approach is not the best for rare codes, multiple specialties or frequent system updates. Corti’s model, based on targeted LLMs, rethinks medical coding as a reasoning process using four agents that mimic the work of human coders. The agents identify evidence, reason through hierarchies, validate against guidelines and reconcile ambiguity, the company said.
The need for clear coding has never been greater, Maaløe told Fierce Healthcare, because with the advent of ambient scribes, unstructured clinical notes are getting longer. That means more potential missed opportunities to code. Though the main use case for Corti is revenue cycle management, the platform can also track indications of fraud, per Maaløe.
Corti is compliant with HIPAA and GDPR privacy standards. Maaløe is finding that U.S. customers are requesting Corti’s product with GDPR-level compliance amid lagging privacy regulations in the U.S.
Corti began in 2016 as a research company trying to understand how to stream a large language model in real time to reduce administrative burden and augment clinician workflow. It has raised $100 million to date.
Steve West, managing director of Healthliant Ventures and Tanner Health, said in an announcement that Corti’s methodology “is the most promising approach to medical coding we’ve seen. We've been co-developing with Corti because we believe specialized AI infrastructure is how this problem gets solved—and we're excited to see it move into production.”