Medical-Specialized Foundation Model ALLM.H (Areum.H)

Acryl has developed its own proprietary industry-specialized foundation model family, "ALLM (Acryl LLM)." Its first deliverable—the medical-specialized model "ALLM.H (Areum.H)"—achieved global SOTA (State-of-the-Art) status in April 2026.

1. Proven, Industry-Leading Performance (Global SOTA)

• Achievement: Scored 96.78% on the Korean medical AI evaluation benchmark KorMedMCQA — ranked #1 globally
• Benchmark Comparison: Outperformed leading commercial models from major tech companies (Claude Opus 4: 96.55% / Gemini 2.5 Pro: 90.8% / GPT-5.1: 90.11%)
• Value: Validated against the Korean Medical Licensing Examination (KMLE), proving a level of reliability suitable for immediate clinical deployment

2. Reduced Deployment Costs (Lightweight Architecture)

• Architecture: Delivers top-tier performance with a lightweight 31B model (fine-tuned on Google Gemma 4), without relying on ultra-large 72B+ models
• Value: Optimized model size dramatically reduces the cost of building and maintaining large-scale GPU server infrastructure at medical institutions

3. Robust Data Security (Optimized for On-Premise Deployment)

• Environment: The lightweight design enables easy standalone deployment within a hospital's internal network (on-premise)
• Value: Operates without external cloud communication, fundamentally eliminating the risk of leaking sensitive medical data and patient personal information

4. Clinical Validation and Service Scalability

• References: Clinical validation underway at major tertiary hospitals including Yonsei University Health System and Kyungpook National University Hospital (linked with Dr. Answer 3.0 and K-ARPA)
• Scalability: Anchor model can be branched into department-specific specialized models for each medical specialty
• Ecosystem: Through integration with Acryl's "JONATHAN AGENTBASE," capabilities extend beyond simple Q&A to function as a practical "medical agent" for real-world clinical workflows

5. A Model Hospitals Can Actually Deploy

• On-Premise Support for Air-Gapped Environments: Operates entirely on the hospital's internal infrastructure — no patient information is transmitted to external clouds (reported April 2026).
• National R&D Program-Based Validation: Real-world validation is being conducted at partner hospitals including Yonsei University Health System and Kyungpook National University Hospital, under the Ministry of Health and Welfare and Ministry of Science and ICT programs "Dr. Answer 3.0" and "K-ARPA."
• Full-Stack Medical AI: Beyond the model itself (Areum.H), ALLM.H integrates with NADIA-ANE, the AI-Native EMR onboarding platform launched on May 12, 2026, automating data standardization and onboarding from day one of deployment.
• Government Program Selections: On April 16, 2026, selected for the Korea Health Information Service's "Medical AI Data Utilization Voucher Support Program" and the "Multi-Institutional Emergency Data Integrated AI Development" program.

6. Key Use Cases

• Clinical Decision Support (CDS): Assisting clinicians with diagnostic and prescription decision-making.
• Medical Data Analysis: Structuring and analyzing unstructured medical text and EMR data.
• Specialty-Specific Consultation Support: Supporting consultations and clinical documentation grounded in department-specific domain knowledge.