Doctor Evidence Search Tool “Evidence Finder” Adds Sakana Namazu

Sakana AI’s Japanese-tuned LLM, Sakana Namazu, is being plugged into Evidence Finder, a doctor-facing evidence search tool from Iris. The service pulls papers from sources like PubMed, shows where its answers come from, and includes a Verify function to check whether cited papers really exist.

That workflow matters because doctors are short on time and the medical literature never stops moving. New papers, guidelines, comparisons — all of it has to be searched, read, and weighed while the clinician is still doing the actual job.

The setup here splits the work between Iris and Sakana AI. Iris’ own search and verification algorithms pick the relevant papers and handle the evidence side; Namazu takes over the part that compares, combines, and turns that material into an answer. Sakana says that made the model fit naturally into a path that gets physicians to primary sources quickly.

The company says the decision was driven by Namazu’s Japanese-language ability and its performance in medicine. An evaluation model using Namazu on Evidence Finder scored 96.4% on the 120th Japanese medical licensing exam in February 2026. Iris says that is the highest public score among domestic foundation models designated by Japan’s Ministry of Economy, Trade and Industry GENIAC, based on materials available as of September 1, 2026.

Sakana also argues that the result is not just about raw knowledge. Medical questions need domain knowledge, reasoning, and the ability to understand Japanese intent clearly enough to present it in a form doctors can check. The company says Namazu keeps the base model’s math, general reasoning, and coding ability on major benchmarks, while improving Japanese instruction following, translation, and Japan-specific context handling.

There’s a broader bet underneath all of this. Sakana wants to make strong open models usable in Japan without wrecking what made them strong in the first place, and it’s also pushing for inference to stay domestic. The immediate win is a better doctor tool; the larger pitch is that “Japanese AI” should mean more than fluent Japanese.