Healthcare data and medical knowledge are vast and complex, while healthcare insurance rules are highly intricate. Both public medical insurance and commercial health insurance are high-value scenarios closely tied to national welfare, fund security, public interest, and people’s immediate rights and benefits. They require the highest standards of accuracy, compliance, security, and traceability.

How can AI truly understand complex medical records, process massive volumes of documents, and support medical insurance fund supervision as well as commercial health insurance risk control?

Today, Unisound presents its answer: the official launch of U1-InsureMed, a large model designed for medical insurance intelligence.

Built on Unisound’s self-developed general-purpose large model foundation and deeply integrated with hundreds of millions of real-world clinical data records, U1-InsureMed connects end-to-end capabilities across the medical insurance value chain. It empowers both public medical insurance scenarios, such as compliance supervision and fund security, and commercial insurance scenarios, such as risk identification and refined claims management. With “one model empowering two domains,” it accelerates the digital and intelligent transformation of the healthcare security system.

The model is now fully available on Unisound’s Token Hub large model service platform, offering standardized APIs for one-click integration, on-demand invocation, and token-based billing. This marks a major step forward in scaling AI applications across the insurance sector.

API access: https://maas.unisound.com/

1. Deep Industry Focus: Reconstructing Core Capabilities for Intelligent Insurance from the Ground Up

U1-InsureMed is purpose-built for vertical healthcare insurance scenarios. Its technical foundation is derived from Unisound’s self-developed general-purpose large model and has been comprehensively upgraded to meet the rigorous requirements of high-value applications.

Foundation support

The model is built on Unisound’s self-developed multimodal general-purpose large model and adopts a three-stage training paradigm: “general cognitive foundation + vertical supervised fine-tuning + preference learning.” It also establishes a three-in-one alignment mechanism involving medical insurance experts, clinicians, and auditors, ensuring that model outputs are not only accurate but also aligned with regulatory ethics and business standards.

Business capabilities

Centering on the full process of public medical insurance and commercial insurance, the model provides capabilities including policy Q&A, intelligent auditing, compliance supervision, risk identification, and claims assistance. Through multimodal semantic understanding and medical-domain parsing capabilities, it can truly “understand medical records, read documents, and apply rules accurately,” significantly improving the efficiency and precision of insurance operations.

Data updates

The model is supported by an industry-grade knowledge base system that enables real-time retrieval and dynamic updates of the latest public and commercial insurance policies and regulations. This provides continuous and authoritative data support, ensuring that business decisions and rule execution remain aligned with the latest regulatory requirements.

In real-world business scenario testing, U1-InsureMed achieved significant improvements across multiple capabilities compared with models of a similar parameter scale. Its medical capability, medical insurance knowledge understanding, coding alignment, and business processing capabilities were all enhanced. Specifically, its medical insurance policy Q&A capability improved by 12.6%, enabling more accurate understanding of complex rules. Its medical insurance coding alignment accuracy improved by 23.4%, significantly strengthening coding consistency and standardization. Its medical insurance business processing capability improved by 6.5%, making medical record compliance judgment more efficient and reliable. Its foundational medical capability improved by 5.4%, further enhancing medical understanding and reasoning. Overall, the model has built measurable performance advantages in key business processes, effectively supporting the implementation of medical insurance supervision and commercial insurance risk control scenarios.

2. Deep Deployment Across Two Scenarios: Empowering High-Quality Development of Public and Commercial Insurance

Medical insurance fund supervision is a fundamental safeguard for people’s livelihoods, while commercial insurance risk control is a key pillar for stable industry operations and the protection of user rights. Both require AI to be highly trustworthy, accurate, secure, and compliant.

With deep medical semantic understanding, precise parsing of multimodal documents, and dynamic policy knowledge evolution, U1-InsureMed connects the critical pathway from underlying data cognition to upper-level business decision-making. Based on this foundation, the model targets two core areas: medical insurance fund supervision and commercial insurance risk control. Driven by both technology and scenario implementation, it is reshaping the operating paradigm of the industry.

Public Medical Insurance: Advancing Regulatory Models and Strengthening Fund Security

In medical insurance fund supervision, U1-InsureMed helps the industry move from traditional “rule-based review” toward a new era of intelligent “cognitive supervision.”

The model can directly process massive volumes of unstructured data, including electronic medical records and examination reports. It can independently read full-process diagnosis and treatment information, cross-verify it with medical expense lists, and efficiently identify various violations, such as medication beyond approved limits, duplicate charges, excessive diagnosis and treatment, and DRG upcoding. It can also automatically identify and match rules in complex scenarios such as chronic and special diseases, comprehensively improving the automation and standardization of medical insurance auditing.

Case 1: Deeply understanding medical records to identify medication violations beyond approved limits

In medical insurance violation audits, certain drugs have clearly defined reimbursement scopes. For example, benazepril hydrochloride tablets are reimbursable only for specific indications, including hypertension, congestive heart failure, and progressive chronic renal insufficiency. By automatically parsing patient medical records, the system can determine whether the patient’s previous diagnoses and current hospitalization records reflect the approved indications, thereby identifying improper medication claims.

Case 2: Determining DRG grouping based on medical records to detect upcoding violations

In DRG/DIP payment supervision scenarios, the model understands medical record content and combines it with medical insurance DRG grouping rules to determine the correct case grouping. It can distinguish between similar disease groups and identify behaviors that attempt to obtain medical insurance funds through higher-coded groupings.

Case 3: Intelligent report interpretation to support health assistant capabilities

As medical insurance services continue to evolve toward health management, medical insurance authorities have growing needs for health education, chronic disease management, and proactive health intervention. Relying on multimodal understanding, the model can intelligently parse physical examination reports, laboratory and imaging reports, and outpatient medical records, enabling intelligent report interpretation and health consultation services. This helps medical insurance institutions extend from traditional payment services to broader health services.

In early 2026, Unisound won the bid for China’s first provincial-level vertical medical insurance large model project: the Jiangsu Provincial Medical Insurance Large Model and Agent Application Project. On the regulatory side, the project builds a 24/7 “AI auditor” and establishes a full-lifecycle intelligent control system. On the public service side, it launches an intelligent health service agent, creating a provincial-level demonstration model driven by both fund supervision and public services. This provides replicable and scalable practical experience for the digital transformation of medical insurance nationwide.

Commercial Insurance: Improving Cost Control, Efficiency, and Service Experience

In the commercial insurance sector, Unisound has deepened cooperation with leading insurance groups. Through its medical insurance large model, it has significantly upgraded claims risk control, raising the cost control rate to approximately 20% and enabling billion-yuan-level incremental cost management across more than 2.6 million claims orders. This effectively helps insurers reduce costs and improve efficiency.

For small and medium-sized insurance companies, Unisound has launched an innovative “pay-per-order” flexible cooperation model, building an intelligent risk control service system that covers customers across different levels and lowering the threshold for digital and intelligent transformation.

Case 1: Critical illness determination

In commercial health insurance claims, critical illness determination is a core and complex review process involving large volumes of medical records, examination results, and disease definition clauses. The model can analyze electronic medical records, discharge summaries, pathology reports, imaging examinations, and surgical records. It deeply understands the disease diagnosis process and combines it with the Critical Illness Insurance Disease Definition Standard and relevant product clauses to automatically complete critical illness determination and risk review.

Case 2: Pre-existing condition identification

In commercial insurance underwriting and claims scenarios, identifying pre-existing conditions is an important part of risk assessment. It directly affects liability determination, claim rationality, and fraud risk control. Based on outpatient records, inpatient records, examination reports, and medication records, the model can conduct in-depth analysis of a policyholder’s historical health status and automatically identify potential pre-existing disease risks.

Case 3: Auto insurance medical review to identify unrelated medical charges

In commercial auto insurance claims, medical expense review is essential for cost control and risk identification. It directly affects claim accuracy and payout amounts. The model can analyze outpatient records, inpatient records, examination reports, and expense lists to intelligently determine the relationship between accident-related injuries and medical services. It can automatically identify examinations, treatments, medications, and consumables unrelated to the traffic accident injury.

For complex scenarios such as hospitalization under the pretext of accident injury, unrelated treatment bundled into claims, and chronic disease medication unrelated to the accident, the model performs cross-validation using injury details, treatment pathways, and medical knowledge. It helps auditors quickly detect abnormal charging risks, improving auto insurance claims review efficiency and cost control capabilities.

3. U1-OCR-Med: Breaking Industry Bottlenecks and Building a Technical Moat for Medical Document Processing

Medical insurance and commercial insurance scenarios involve complex processes and a wide variety of paper documents. Most general-purpose large models on the market still struggle with medical documents that feature disordered layouts, seal occlusion, blurred image quality, and low-resolution scans. Common problems include inaccurate recognition, logical discontinuity, and slow scenario adaptation.

U1-InsureMed directly addresses the pain points of public and commercial insurance scenarios by working in synergy with Unisound’s self-developed U1-OCR-Med model. U1-OCR-Med focuses on intelligent recognition, information extraction, and structured parsing of various paper documents. U1-InsureMed then undertakes subsequent business rule verification, audit judgment, and full-process business handling. Together, the two models create comprehensive technical advantages across medical and insurance scenarios, effectively solving industry challenges such as difficult paper document processing and low business circulation efficiency.

Multi-dimensional leading performance

In key medical document tasks such as document classification, bill extraction, medical record parsing, and certificate recognition, the model’s accuracy leads general-purpose large models. All four key capabilities exceed 95%, and the model demonstrates strong cross-domain zero-shot generalization, enabling flexible adaptation to diverse scenarios including public medical insurance and commercial insurance.

Innovative architecture

The model adopts an innovative decoupled structure-aware architecture, allowing it to learn both text semantics and layout features simultaneously. This effectively solves challenges such as disordered medical document layouts, low-quality scanned copies, and seal occlusion. To address cross-page medical records and fragmented information, the model introduces a coordinate-enhanced multi-page joint attention mechanism. Through both spatial and temporal encoding, it enables accurate verification of cross-page related information and resolves information loss and logical discontinuity in ultra-long text processing.

Agile business adaptation

The model supports schema-level flexible field configuration. Business users can quickly adapt to policy adjustments across different provinces and cities without retraining the model. This greatly reduces the maintenance cost of long-tail documents and enables a leap from “technical parameter tuning” to “business self-configuration.”

4. Building a Commercial Ecosystem and Accelerating the Digital Transformation of Healthcare Security

The value of technology is ultimately proven through commercial implementation. Unisound has built a complete healthcare AI business closed loop. Leveraging nearly 450 partner hospitals nationwide, with approximately 35% coverage among A++ hospitals, Unisound continuously feeds hundreds of millions of real-world clinical medical records back into model training. This creates a positive business flywheel: top-tier hospital data accumulation, industry-leading medical and medical insurance large model refinement, and full-scenario expansion across public medical insurance, commercial insurance, and regional medical administration. Together, these factors drive sustained high-speed business growth.

For AI companies, business value equals intelligent density — stronger intelligence from smaller models — multiplied by token value, rather than token quantity. U1-InsureMed moves away from the traditional approach of blindly increasing parameter size and invocation volume. Instead, it focuses on deeply mining high-value tokens. Through efficient data training, precise routing, and dynamic quantization compression, the model delivers higher-density business intelligence output under the same computing power.

On the Token Hub platform, Unisound adheres to “billing based on effective business value,” ensuring that every token invocation directly serves key decision-making points in medical insurance auditing and commercial insurance risk control. This multiplier effect of “high-density intelligence × high-value tokens” not only greatly reduces computing and integration costs for insurers and medical insurance bureaus, but also significantly improves AI return on investment.

Currently, Unisound’s insurance SaaS service has been fully launched. Centered on “U1-InsureMed + medical OCR + intelligent risk control,” it provides hospitals, medical insurance bureaus, and insurance companies with integrated cloud-based services covering intelligent auditing, risk management, fast claims processing, and fund supervision. Through open APIs and lightweight billing on the Token Hub platform, Unisound greatly lowers the threshold for AI application implementation across the industry, helping partners quickly build dedicated intelligent agents and jointly develop a digital healthcare insurance ecosystem.

Looking ahead, Unisound will continue to deepen its focus on the healthcare and insurance verticals, continuously iterate its model technologies, and upgrade its intelligent service capabilities. With a more open ecosystem and more professional solutions, Unisound aims to support the digital and intelligent transformation of China’s healthcare security system, helping it move toward a new stage of development that is more efficient, precise, and inclusive.