StanfordAIMI/CheXOne

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Jan 21, 2026License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

StanfordAIMI/CheXOne is a 3 billion parameter vision-language foundation model, post-trained on Qwen2.5VL-3B-Instruct, designed for chest X-ray interpretation. It features a 32768-token context length and excels at producing explicit reasoning traces alongside answers, supporting multi-task applications like Visual Question Answering, Report Generation, and Visual Grounding. This model is specifically optimized for drafting resident-level radiology reports, matching or outperforming human residents in 55% of cases.

Loading preview...

Overview

StanfordAIMI/CheXOne is a 3 billion parameter vision-language foundation model, post-trained from Qwen2.5VL-3B-Instruct, specifically developed for chest X-ray interpretation. It integrates advanced reasoning capabilities to provide explicit, step-by-step traces alongside its final answers, enhancing transparency and reliability in medical diagnostics. The model supports a substantial context length of 32768 tokens, enabling comprehensive analysis of medical images and associated textual queries.

Key Capabilities

  • Reasoning Capability: Generates explicit reasoning traces, detailing the thought process behind its conclusions.
  • Multi-Task Support: Handles diverse tasks including Visual Question Answering (VQA), Report Generation, and Visual Grounding for chest X-rays.
  • Resident-Level Report Drafting: Demonstrates high proficiency in drafting radiology reports, matching or exceeding the quality of resident-drafted reports in 55% of evaluations.
  • Dual Inference Modes: Offers a Reasoning Mode for higher performance with detailed traces and an Instruct Mode for faster inference without explicit reasoning.

Good For

  • Medical Imaging Analysis: Ideal for automated interpretation of chest X-rays, providing findings and diagnoses.
  • Clinical Decision Support: Assists radiologists and clinicians by generating preliminary reports and offering reasoning for its conclusions.
  • Research and Development: Useful for researchers exploring advanced vision-language models in the medical domain, particularly those requiring explainable AI outputs.
  • Educational Purposes: Can serve as a tool for training and evaluating medical students and residents in radiology interpretation.

Finetunes
2 models