Research
Notes.One at a time.
Numbers and releases.
- November 2026Arxo Materia EHRSQL 0.1A 4B model reaches 67.5% on EHRSQL in MedHELM, and a 0.8B model 66.8%. The best result on the MedHELM leaderboard is 63.2%. The models, their answers and scores are public.Release
- November 2026Arxo Materia MedCalc 0.1A 4B model reaches 90.2% on MedCalc-Bench Verified with Arxo, and 82.2% on its own. A 0.8B model reaches 70.9%. The standalone models, their answers and scores are public.Release
- November 2026CalculemusMachines have learned to speak. Fluency is one half of intelligence. Arxo is the other half.Note