HuatuoGPT Medical AI Ecosystem

Open medical language, vision, reasoning, data, and deployment infrastructure for clinical AI.

Medical AI ecosystem
AI for Healthcare Clinical reasoning Medical vision-language Open data Deployment
HuatuoGPT HuatuoGPT-o1 HuatuoGPT-Vision Reasoning data
HuatuoGPT medical AI overview

HuatuoGPT is the lab's flagship medical AI program: an open ecosystem that moves from Chinese medical QA data to doctor-facing chat models, medical complex reasoning, multimodal medical vision, and real-world triage and pre-consultation systems.

The project is not a single model release. It is a stack: medical data construction, supervised medical adaptation, verifier-style reasoning problems, visual medical knowledge injection, evaluation, and deployment loops that expose what medical AI can and cannot safely do.

Research Storyline

Data
Start with Chinese medical QA

Huatuo-26M and the early HuatuoGPT work created a reusable Chinese medical question-answering substrate for medical instruction tuning.

Adapt
Turn general LLMs into medical assistants

HuatuoGPT-II simplifies medical adaptation with a one-stage recipe and public checkpoints, making the training pipeline easier to reproduce.

Reason
Move from chat to complex diagnosis

HuatuoGPT-o1 introduces medical complex reasoning data and models, pushing beyond fluent consultation toward step-by-step clinical reasoning.

See
Add visual medical knowledge

HuatuoGPT-Vision and PubMedVision connect medical images, reports, and dialogue so the ecosystem can handle multimodal clinical questions.

Deploy
Close the loop with real workflows

Longgang triage and pre-consultation deployment turns evaluation pressure back into product and research questions.

What It Builds

Open medical LLMs

HuatuoGPT and HuatuoGPT-II adapt general LLMs to medical consultation, QA, and instruction-following with domain-specific supervision and evaluation.

Complex medical reasoning

HuatuoGPT-o1 introduces verifiable medical problems and reasoning SFT data, targeting multi-step diagnosis, treatment reasoning, and medical decision support.

Medical multimodality

HuatuoGPT-Vision and PubMedVision inject medical visual knowledge into multimodal LLMs for image-grounded medical dialogue and visual QA.

Deployment path

HuatuoGPT-powered intelligent triage and pre-consultation systems have been used in Longgang, Shenzhen public hospital and community health settings.

Project Gallery

HuatuoGPT ecosystem overview
The ecosystem view: data, medical LLMs, reasoning, multimodality, and deployment form one pipeline.
HuatuoGPT-II automatic evaluation
HuatuoGPT-II focuses on a simpler medical adaptation recipe and reproducible evaluation.
HuatuoGPT Longgang deployment
Deployment in Longgang turns model research into triage and pre-consultation workflow feedback.

Representative Work

Huatuo-26M

A large Chinese medical QA dataset that seeded the early HuatuoGPT training pipeline and made Chinese medical instruction data reusable.

Code and data
HuatuoGPT-o1

Medical complex reasoning models trained with medical-o1-reasoning-SFT and verifiable medical problem data.

Model
HuatuoGPT-Vision

A multimodal medical model line connected to PubMedVision, a large image-text medical visual instruction dataset.

Model

Paper Trail

Dataset
Huatuo-26M, a Large-scale Chinese Medical QA Dataset

Builds the data foundation for Chinese medical instruction tuning and evaluation.

Code and data
Model
HuatuoGPT, towards Taming Language Model to Be a Doctor

Turns the data foundation into an open medical dialogue model and establishes the first HuatuoGPT model family.

Repository
Adapt
HuatuoGPT-II, One-stage Training for Medical Adaption of LLMs

Simplifies the adaptation pipeline and releases public checkpoints for downstream medical LLM research.

Repository
Reasoning
HuatuoGPT-o1: Towards Medical Complex Reasoning with LLMs

Adds verifier-style medical reasoning data and models to make medical outputs more structured and testable.

Repository
Vision
HuatuoGPT-Vision, Towards Injecting Medical Visual Knowledge into Multimodal LLMs at Scale

Extends the ecosystem from text-only medical QA to image-grounded medical dialogue and multimodal knowledge.

Repository

Why It Matters

  • It turns medical AI from a demo into a reproducible stack of datasets, models, benchmarks, and deployment artifacts.
  • It connects text-only consultation, visual medical understanding, and explicit reasoning rather than treating them as isolated model families.
  • It provides open entry points for the community: GitHub repositories, Hugging Face checkpoints, and medical reasoning datasets.