AxonMed-7B: Sovereign Clinical Foundation Model via Multi-Stage Curriculum Scaling and Medical o1 Reasoning
Year: 2026
Type: preprint
Abstract: AxonMed-7B is an open-weights, clinical-grade 7-billion parameter medical foundation model engineered by SaathiMed Research Lab. Built upon Qwen2.5-7B-Instruct and trained across 249,000+ clinical vignettes (AIIMS/NEET-PG MedMCQA, USMLE Step 1/2/3 MedQA, NBME Self-Assessments, PubMedQA, and 19,704 Medical o1 reasoning trajectories), AxonMed-7B achieves 61.25% accuracy on MedMCQA and 75.40% on PubMedQA via EleutherAI lm-evaluation-harness, outperforming BioMistral-7B (51.80%), Meditron-7B (52.30%), Gemma-7B (52.80%), Meta LLaMA-3-8B (54.10%), and OpenAI GPT-3.5 Turbo (60.50% / 75.20%). Model weights and reproducibility scripts are openly released at https://huggingface.co/Saathimed/AxonMed-7B.
Author: Kumar, Amrendra
Institution: Indian Institute of Technology Patna
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7225796365
Cites:
Cited by:
Related to: