Sakana AI’s Namazu Model Powers Aillis’s Evidence Finder Search Tool for Doctors
Sakana AI said on October 9 that Sakana Namazu, its large language model tuned for Japanese, has been adopted by Evidence Finder, an evidence search tool for doctors offered by the Tokyo medical AI company Aillis, AI Watch (Impress) reported. The tool takes a doctor’s question, searches databases such as PubMed and writes an answer that shows where each claim comes from.
The two companies have split the work. Aillis’s own search and verification algorithms pick the relevant papers for each question and compare and integrate them, while Sakana Namazu handles the step that generates the answer. A built-in function called Verify automatically checks that the papers the AI cites actually exist, a safeguard against invented references, and the service is designed to get doctors to the primary literature quickly. Aillis first announced the agreement to use Sakana AI’s model on September 14, saying the detailed algorithm is patent-pending and undisclosed.
The choice followed testing. An evaluation version of Evidence Finder running on Sakana Namazu sat the 120th National Medical Licensing Examination, held in February, and reached a 96.4% accuracy rate, scoring 482 out of 500. Aillis says that is the highest score it could confirm from public data among the domestic foundation models designated under GENIAC, the generative AI programme of the Ministry of Economy, Trade and Industry. Sakana AI said that result, together with a qualitative review of real answers that found them useful in practice, led to the adoption.
What it means in practice
Exam scores make headlines, but the details matter. Aillis notes that the commercial Evidence Finder answers text questions only and does not handle the image questions that appear on the licensing exam; the test lifted the restrictions placed on the product model to measure its underlying capability. The company also states that the service is not a medical device and is not meant to diagnose, treat or prevent disease in individual patients. For doctors, the practical value lies in faster, cited literature searches in Japanese rather than in automated clinical decisions.
The deal gives Sakana AI a showcase in a regulated field, following the release of its Fugu Max and Fugu Ultra v2 orchestration models and the appointment of Jürgen Schmidhuber as chief scientific advisor. It also fits the government’s push for domestic, or sovereign, AI in healthcare, which Aillis ties to the Basic Policy on Economic and Fiscal Management and Reform 2026. Hospitals are testing AI in other ways too, as seen in shared playbooks for imaging triage.
What to watch: whether Aillis publishes accuracy data from real clinical use, whether other medical tools choose GENIAC-backed domestic models, and how Evidence Finder handles the image-based questions it currently leaves out.