AI Sets New Accuracy Standard in Medical Coding with Minset's mCoder

BOSTON, MA – May 14, 2026 – A new high-water mark has been set in the complex field of automated medical coding. Boston-based Minset announced today that its autonomous AI system, mCoder, has achieved state-of-the-art performance on a leading public benchmark, outperforming all previously published peer-reviewed results and signaling a significant leap toward fully autonomous systems in healthcare revenue cycle management.

The achievement was recorded on MDACE (MIMIC Documents Annotated with Code Evidence), a rigorous benchmark introduced in 2023 by researchers from 3M Health Information Systems and Carnegie Mellon University. The benchmark is specifically designed to test an AI's ability not just to assign medical codes, but to ground them in specific evidence within complex clinical notes, a critical requirement for real-world auditing and compliance.

A New Benchmark for AI in Healthcare's Financial Backbone

Medical coding is the translation of healthcare diagnoses, procedures, services, and equipment into universal alphanumeric codes, forming the foundation of billing and reimbursement. It is a notoriously complex, labor-intensive process that requires deep clinical knowledge and attention to detail. Errors or inefficiencies can lead to claim denials, delayed payments, and significant administrative costs for healthcare providers.

MDACE was created to provide a standardized, public yardstick for the next generation of AI-powered coding tools. It is widely respected for its difficulty, using real-world hospital admission data that includes a mix of discharge summaries, physician notes, and procedural reports for both inpatient and professional-fee scenarios.

Crucially, the benchmark evaluates AI models on two distinct levels of difficulty:

• 1K-constrained setting: A simplified test where the AI chooses from the roughly 1,000 codes present in the MDACE dataset.
• Full-label setting: A far more challenging and clinically realistic task where the AI must select the correct codes from the entire ICD-10 space, which contains over 70,000 possibilities.

Minset's mCoder surpassed all known prior results on both settings, but its performance on the full-label test is particularly noteworthy. This demonstrates a capability to operate at the scale and complexity required in actual production environments, a hurdle where many systems falter.

"We believe this represents an important step toward fully autonomous, production-grade coding systems that can operate reliably at scale across inpatient and professional-fee workflows, with the generality required for broader real-world coding environments," said Matt Scott, CTO of Minset.

Raising the Bar in a Competitive Field

Minset's announcement lands in an increasingly competitive landscape for healthcare AI. Companies like CodaMetrix and Fathom are established players, and newer entrants like Corti have also reported results on the MDACE benchmark, cementing its status as a key battleground for demonstrating technological prowess. The race is on to automate a function that costs a mid-sized hospital millions annually in labor and rework.

While leading AI platforms have reported accuracy rates between 82% and 97% depending on the complexity of the case, recent studies have shown that general-purpose large language models (LLMs) still struggle significantly with the nuances of medical coding. One April 2024 study found that even the most advanced general LLMs had exact-match accuracy rates below 34% for ICD-10 codes, highlighting the need for specialized, purpose-built AI like mCoder.

Minset’s achievement suggests its system has crossed a critical threshold in specialized understanding, moving beyond simple pattern matching to a more robust interpretation of clinical documentation. The company, founded by former leaders from Microsoft Research, Google, Optum, and Salesforce, is leveraging this deep technical expertise to tackle one of healthcare's most persistent administrative challenges.

Beyond Accuracy: The Push for Auditable Automation

The significance of the MDACE benchmark extends beyond a simple accuracy score. Its emphasis on explainability—requiring the AI to pinpoint the exact text that supports each code—addresses a fundamental need for healthcare organizations: auditability.

In revenue cycle operations, a code without evidence is a compliance risk. Coders, auditors, and clinicians must be able to trace every billing decision back to the patient's chart. Minset asserts that mCoder’s superior performance on this benchmark demonstrates an advanced ability to create a transparent and defensible audit trail for every code assigned.

This capability is a core component of Minset's broader strategy. The mCoder system is part of an integrated intelligent revenue cycle platform that also includes mDenials for automating the management of denied claims and m360 for patient engagement. According to the company, these modules operate on a shared reasoning framework, creating a closed-loop system where insights from denials can continuously improve the accuracy of initial coding, and vice versa. This unified approach aims to replace the fragmented, manual processes that currently plague hospital finance departments.

The Evolving Role of the Medical Coder

The rise of powerful autonomous systems inevitably raises questions about the future for the human workforce. For the tens of thousands of medical coding professionals, the advent of AI like mCoder signals not an elimination of their role, but a profound evolution.

Industry experts suggest that AI will automate the high-volume, repetitive coding tasks, freeing human professionals to focus on more complex and valuable work. Instead of manually poring over every chart, coders will increasingly act as auditors, exception handlers, and quality assurance experts. The model is shifting towards human-AI collaboration, where the technology handles the bulk of the work with a certain confidence score, and any ambiguous or low-confidence cases are automatically routed to a human expert for review.

This shift allows human coders to apply their expertise to the most challenging cases, manage compliance, and defend coding decisions during audits—tasks that require critical thinking and nuanced judgment that AI has yet to master. The goal is to elevate the profession, reduce burnout, and allow healthcare organizations to deploy their skilled human capital more strategically.

As Minset begins to evaluate mCoder in production environments with select partners, the healthcare industry will be watching closely. The transition from a benchmark champion to a proven, enterprise-scale solution will be the ultimate test of whether autonomous coding is ready to fundamentally reshape how the business of healthcare gets done.