Mitihoon – SCB 10X, the disruptive technology investment arm of SCBX Group, announced a research partnership with the Faculty of Medicine Siriraj Hospital, Mahidol University, to advance medical AI through cross-disciplinary collaboration. This partnership integrates clinical expertise, medical data, and real-world problem insights from Siriraj’s SiData+ with the advanced AI expertise from Typhoon by SCB 10X.
This collaboration has led to the research titled “On the Robustness of Answer Formats in Medical Reasoning Models,” which reflects a new paradigm in medical AI development, moving beyond single-answer outputs toward reasoning models that “think before responding” and place emphasis on answer formats that align with real-world clinical practice. This approach is designed to support medical professionals in making more comprehensive and safer clinical decisions.
One of the project’s key outcomes is Typhoon-Si-Med-Thinking-4B, a compact medical AI model designed to act as a clinical reasoning assistant—not a replacement for physicians. Unlike general large language models (LLMs) that produce a single answer, the model generates ranked diagnostic lists that mirror real-world differential diagnosis workflows. The model is the first to be specifically designed and trained to generate ranked diagnostic lists, closely mirroring physicians’ real-world differential diagnosis processes. This innovation helps reduce bias and enhances clinical reasoning support compared with general LLMs, which typically produce only a single answer.
Preliminary evaluations indicate that, despite having significantly smaller in size than large commercial language models from overseas, Typhoon-Si-Med-Thinking-4B can generate ranked diagnostic outputs with greater accuracy. This demonstrates the potential of specialized small-scale AI models that deliver high performance while remaining cost-effective and accessible. The model is currently in an early research stage and requires further testing with medical professionals for real-world use. The primary objective of this release is to disseminate knowledge openly, enabling external researchers to access and build upon this work for the greatest benefit to Thailand’s public health system and society at large.
Technical details of the model and research are available at opentyphoon.ai. Further insights will be presented at SiData+ Conference 2026, held on Friday, 23 January 2026, at Veekit Veeranuwat Auditorium, Siriraj Hospital. The session, titled “Single Answer Is Not Enough: On Generating Ranked Lists with Medical Reasoning Models,” is hosted in collaboration with the Faculty of Medicine Siriraj Hospital.y. Registration is available at https://www2.si.mahidol.ac.th/sirirajconference/event/sm/content/3528/TH/34276
ติดตามช่องทางมิติหุ้นเพื่อรับข่าวสารตลาดทุนได้ตามลิงค์ด้านล่าง
Web : https://www.mitihoon.com/
Facebook : https://www.facebook.com/mitihoon
Youtube : https://www.youtube.com/@mitihoonofficial7770
Tiktok : www.tiktok.com/@mitihoon
