# FreedomIntelligence releases HuatuoGPT-3 multimodal model on Hugging Face

Digest AI · Generative AI & Models · published 2026-09-24T19:20:54Z

Canonical: https://digestai.news/story/freedomintelligence-releases-huatuogpt-3-multimodal-model-on-hugging-f

## Summary

The model builds on **Qwen3.8-27B** with a reinforcement-learning method called **One-stage Policy Optimization (OnePO)**. It requires a *thinking mode* during inference, generating reasoning before final answers, and supports image-text-to-text tasks.

The model’s training code, medical reinforcement-learning dataset, and an 8B rubric grader are also open. Users can deploy it via **vLLM** or **SGLang** or run it locally with Docker. Example prompts include medical queries like *‘What are the common causes of chest pain?’* or image descriptions. No performance benchmarks or context window size are provided.

## Key points

- HuatuoGPT-3 uses OnePO to adapt language models to medical tasks without supervised fine-tuning
- Model requires *enablethinking=True* for reasoning before final answers, with a </think> delimiter
- Open-source on Hugging Face with training data, code, and an 8B rubric grader released

## Why it matters

A medical LLM with open reinforcement-learning training could accelerate research and deployment in healthcare AI, though its performance and context limits are untested.

## Sources

1. [FreedomIntelligence/HuatuoGPT-3-27B · Hugging Face](https://huggingface.co/FreedomIntelligence/HuatuoGPT-3-27B) (huggingface.co, 2026-09-24, primary source)

Part of the developing story: [Open Source AI Infrastructure and Medical Model Expansion](https://digestai.news/thread/ggml-org-merges-hc-ops-into-qwen4exp-graph-in-llama-cpp) (4 stories)

## Cite

Digest AI, "FreedomIntelligence releases HuatuoGPT-3 multimodal model on Hugging Face", 24 September 2026, https://digestai.news/story/freedomintelligence-releases-huatuogpt-3-multimodal-model-on-hugging-f

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/freedomintelligence-releases-huatuogpt-3-multimodal-model-on-hugging-f.json