VARYTEX Develops Kampo-Specialized AI 'KAMPO LLM' - Achieves 97.4% Accuracy on 471 Kampo Specialist Questions, Outperforming Major LLMs

Key facts

  • VARYTEX Develops Kampo-Specialized AI 'KAMPO LLM' - Achieves 97.4% Accuracy on 471 Kampo Specialist Questions, Outperforming Major LLMs
  • VARYTEX Inc., in collaboration with the Japan Society for Oriental Medicine, has developed and evaluated a Kampo-specialized AI, KAMPO LLM. Using the society's core curriculum of 471 questions, KAMPO LLM Pro achieved 97.4% accuracy, outperforming current general-purpose AI models from Google and OpenAI.
  • Source: PR Times
  • Date: June 8, 2026

Direct answer

VARYTEX Inc., in collaboration with the Japan Society for Oriental Medicine, has developed and evaluated a Kampo-specialized AI, KAMPO LLM. Using the society's core curriculum of 471 questions, KAMPO LLM Pro achieved 97.4% accuracy, outperforming current general-purpose AI models from Google and OpenAI.

Citation
VARYTEX Develops Kampo-Specialized AI 'KAMPO LLM' - Achieves 97.4% Accuracy on 471 Kampo Specialist Questions, Outperforming Major LLMs (June 8, 2026), PR Times
Source
PR Times
Date
June 8, 2026
VARYTEX Inc., in collaboration with the Japan Society for Oriental Medicine, has developed and evaluated a Kampo-specialized AI, KAMPO LLM. Using the society's core curriculum of 471 questions, KAMPO LLM Pro achieved 97.4% accuracy, outperforming current general-purpose AI models from Google and OpenAI.

📋 Article Processing Timeline

  • 📰 Published: June 8, 2026 at 21:00
  • 🔍 Collected: June 8, 2026 at 12:21
  • 🤖 AI Analyzed: June 12, 2026 at 14:43 (98h 21m after Collected)
VARYTEX Inc. (Headquarters: Minato-ku, Tokyo; Representative Director: Kiichiro Hirano) has developed and evaluated its proprietary Kampo-specialized AI, KAMPO LLM, as part of a joint research project with the Japan Society for Oriental Medicine. This research was conducted with cooperation from the Society's International Committee from the planning stage. A comprehensive evaluation using the Society's core curriculum for Kampo specialist training, 'Questions and Explanations' (471 questions in total), confirmed that KAMPO LLM Pro achieved an accuracy rate of 97.4% (459/471 questions), surpassing the latest general-purpose AI models offered by Google and OpenAI. We are pleased to announce this research achievement. This accuracy is difficult to achieve with general-purpose AI models alone and demonstrates KAMPO LLM's high performance in the specialized field of Kampo medicine.

■ Evaluation Results: Comparison with Major AI Models Using Kampo Specialist Training Text

In this evaluation, using the Japan Society for Oriental Medicine's core curriculum for Kampo specialist training, we compared KAMPO LLM with major large language models. Results showed high performance in the accuracy of specialized knowledge and interpretive ability.

1. High-Precision AI 'KAMPO LLM Pro' Comparison (All 471 Questions)

Model
Accuracy Rate
Number Correct

KAMPO LLM Pro
97.4%
459/471

Gemini 2.5 Pro (Google)
91.7%
432/471

GPT-5.4 Pro (OpenAI)
91.3%
430/471

o3 (OpenAI)
77.1%
363/471

GPT-4o (OpenAI)
58.8%
277/471

GPT-4.1 (OpenAI)
56.8%
266/468

2. High-Speed, Low-Cost AI 'KAMPO LLM Flash' Comparison (All 471 Questions)

Model
Accuracy Rate
Number Correct

KAMPO LLM Flash
92.1%
434/471

Gemini 2.5 Flash (Google)
80.3%
378/471

GPT-5.4 (OpenAI)
79.2%
373/471

o4-mini (OpenAI)
61.8%
291/471

GPT-4o (OpenAI)
58.8%
277/471

KAMPO LLM Flash, intended for practical use, achieved an accuracy rate of 92.1% while delivering approximately three times the response speed and about 60% cost reduction compared to the Pro version. This precision surpasses not only other companies' lightweight models but also the top-tier general-purpose AI models.

■ Three Core Values Realized by KAMPO LLM

1. Advantage of Proprietary Data Rooted in Clinical Practice

Utilizes real clinical data accumulated through operation of its own DX services and VARYTEX's proprietary Kampo medicine database. It deeply understands 'Japan's unique medical system' that global companies cannot access.

2. Improvement of Answer Accuracy in Specialized Domains

For Kampo-specific concepts where general-purpose AI tends to struggle (determination of 'sho', selection of formulas, etc.), we confirmed that more accurate answers are possible through proprietary data and optimization.

3. Immediate Integration into DX Ecosystem

This AI is scheduled to be integrated into the Kampo business support software 'KAMPO 365' and offered as an enterprise API.

■ Future Outlook: Toward Kampo Medicine Infrastructure

VARYTEX will expand this technology not only for medical professional diagnosis and business support but also into the B2C domain, allowing the general public to access reliable Kampo knowledge. Under the theme 'Digital Transformation of Kampo,' we aim to elevate this 1,500-year-old traditional medicine to a digital-era standard and expand the reach of Kampo medicine worldwide.

KAMPO LLM Product Page: https://kampollm.varytex.co.jp/

■ Dr. Kang Su Kim, Vice President of Matsuo Research Institute, LLC (Outside Director of VARYTEX Inc.)

The true value of technology is determined by collaboration with those who know the field intimately. Having been involved in the AI world for a long time, I find increasing occasions to feel this. The achievement of VARYTEX surpassing the latest models from OpenAI and Google in the Kampo domain is a testament to this.

There are certainly areas where meticulous collaboration with on-site data and experts can directly compete against multi-trillion-yen development investments. Rather, the deeper the expertise and the more field knowledge matters, the more pronounced the advantage of this approach becomes.

I sincerely look forward to the day when KAMPO LLM grows into a global infrastructure for Kampo medicine, allowing more people around the world to access high-quality Kampo wisdom.

■ Hayato Otake, CTO of VARYTEX Inc.

Kampo medicine is a unique medical system unparalleled in the world, having evolved independently in Japan from Chinese traditional medicine. Despite its history of over 1,500 years, its digitalization lags significantly behind other medical fields, and high-quality data specialized in Kampo medicine has not been sufficiently accumulated even by global AI companies like OpenAI and Google.

The current evaluation confirmed that while general-purpose AI models show high accuracy in some areas of Kampo medicine, they have clear limitations in specialized domains. To advance the fusion of Kampo medicine and AI, it is essential to have entities that understand the Kampo field, collaborate with specialists, and accumulate unique data and knowledge.

Through the development and operation of multiple Kampo DX services, VARYTEX has been fulfilling this role. By leveraging insights gained from clinical practice and proprietary data accumulated through the operation of its own services, we have succeeded in developing a Kampo-specialized AI that surpasses general-purpose AI models.

This achievement was made possible thanks to the guidance of the doctors from the planning stage as a joint research with the Japan Society for Oriental Medicine. We are confident that this initiative, in which academic societies and companies collaborate to utilize AI technology in Kampo medicine, is indispensable for the future of Kampo medicine in promoting the standardization and appropriate use of Kampo theory.

Going forward, we will continue to work on improving the accuracy of KAMPO LLM and its application in clinical practice, in collaboration with Kampo specialists including the Japan Society for Oriental Medicine.

■ Acknowledgements

This research was conducted as a joint research project with the Japan Society for Oriental Medicine. We deeply appreciate the cooperation of the members of the Society's International Committee from the planning stage, and the doctors of the Institute of Oriental Medicine at Tokyo Women's Medical University for their cooperation in creating training data.

Part of this technology was researched and developed under the New Product/New Technology Development Support Project of Minato City.

■ About VARYTEX Inc.

Company Name: VARYTEX Inc.

Address: 1F Roppongi Adachi Building, 3-4-24 Roppongi, Minato-ku, Tokyo 106-0032, Japan
Established: March 31, 2021
Capital: 116,917,730 yen
Representative: Kiichiro Hirano, Representative Director CEO & Founder
URL

FAQ

What is KAMPO LLM?

It is an AI model specialized in Kampo medicine developed by VARYTEX, achieving 97.4% accuracy on specialist training texts.

What data is KAMPO LLM trained on?

It utilizes real clinical data accumulated from VARYTEX's own Kampo DX services and their proprietary Kampo medicine database.

Where can I use KAMPO LLM?

It is planned to be integrated into the business support software 'KAMPO 365' and offered as an enterprise API.