Announced details
Specialist medical examination (RAG) accuracy: up to 90.8%
Improvement of up to 10.8 percentage points over the base model in guideline evaluations
JLAC11 code conversion accuracy: up to 80.3%
Case-data organization accuracy: 92.2%
Discharge-summary quality evaluation: 4.748/5 points
Red-teaming: approximately 6,000 cases
More than 50,000 cases in the interactive safety benchmark
Helios was one of the consortium institutions selected to jointly implement this project. No other capital or personnel relationships are stated.
NEDO is a national research and development agency. The document provides no individual details for the other consortium institutions, such as their locations, representatives, capital or year of establishment.
The completion of this project (project period: fiscal 2025 / 1 year) will have no impact on consolidated results for the fiscal year ending December 2026.
The company does not address Contract Value / Offering price.
NEDO, “Project to Promote Research and Development, Verification, and Other Activities to Ensure AI Safety”
| Item | Details |
|---|---|
| Names of Products / Services Covered | Japanese LLM for supporting medical operations, Weblab-MedLLM-GLM-4.7, Weblab-MedLLM-Qwen3-235B-Thinking, Weblab-MedLLM-Qwen3-235B-Instruct, Weblab-MedLLM-gpt-oss-120b, Medical-Qwen3-Swallow-30B-A3B, Medical-Qwen3-Swallow-32B, Medical-Qwen3-Swallow-8B, Medical-GPT-OSS-Swallow-120B, AscleLM1-30B-A10B, AscleLM1-10B |
| Counterparty | New Energy and Industrial Technology Development Organization (NEDO), SAKURA internet Inc., The University of Tokyo, ABEJA, Inc., RIKEN, a national research and development agency, International University of Health and Welfare, Fujita Health University, Fujita Academy, Institute of Science Tokyo, National University Corporation Kyushu University |
| Counterparty overview | NEDO is a national research and development agency. The document provides no individual details for the other consortium institutions, such as their locations, representatives, capital or year of establishment. |
| Relationship with the Company | Helios was one of the consortium institutions selected to jointly implement this project. No other capital or personnel relationships are stated. |
| Launch date | not disclosed (phased social implementation is planned) |
| Applicable area | Support for administrative tasks and document preparation at medical institutions, Automatic conversion of examination names into JLAC11 codes, Automatic organization of case data, Drafting of discharge summaries, Natural-language queries to electronic medical records |
| Quantitative Effects | Specialist medical examination (RAG) accuracy: up to 90.8%, Improvement of up to 10.8 percentage points over the base model in guideline evaluations, JLAC11 code conversion accuracy: up to 80.3%, Case-data organization accuracy: 92.2%, Discharge-summary quality evaluation: 4.748/5 points, Red-teaming: approximately 6,000 cases, More than 50,000 cases in the interactive safety benchmark |
| Impact on Business Performance | The completion of this project (project period: fiscal 2025 / 1 year) will have no impact on consolidated results for the fiscal year ending December 2026. |
| Source | NEDO, “Project to Promote Research and Development, Verification, and Other Activities to Ensure AI Safety” |
Items this document does not state (9)
These are items the company does not state in this document. We do not leave them blank, nor fill them in from other sources or by inference.
The reasons the company gives
The company's own explanation, not something inferred from the numbers.Announcement of development results from NEDO’s “Project to Promote Research and Development, Verification, and Other Activities to Ensure AI Safety.” The project involved developing an LLM, verifying its safety, and testing use cases, with the goal of operating it in an environment where patient information can be managed securely and achieving performance comparable to leading commercial LLMs.