8.1B- Context
- 131Ktokens
- Released
- 1y ago
- Size
- 32Bparameters
Baichuan-M2-32B is a medical reasoning model from Baichuan, post-trained from Qwen2.5-32B and the second medical model the company has released. The maker says a Large Verifier System of patient simulators and multi-dimensional verification, plus multi-stage reinforcement learning, yields a large gain in medical performance while keeping general abilities, and claims 58.5% higher token throughput for its MTP version in single-user scenarios. It is built for real-world medical reasoning work such as clinical decision support, health consultation and medical education.
How it compares
Additional source evidence
- Source model
- baichuan/baichuan-m2-32b
- Source family
- baichuan
- Last updated
- 2025-08-13
- Model context
- 131K
- Model output limit
- 131K
- Open weights
- Yes
- File attachments
- No
- Reasoning
- Yes
- Tool calling
- No
- Temperature control
- Yes
Downloadable weights
Complete source record (JSON)
{
"id": "baichuan/baichuan-m2-32b",
"name": "Baichuan M2 32B",
"description": "Open medical model trained for clinical reasoning and patient interaction with on, off, and auto thinking modes",
"family": "baichuan",
"last_updated": "2025-08-13",
"attachment": false,
"reasoning": true,
"tool_call": false,
"temperature": true,
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 131072,
"output": 131072
},
"release_date": "2025-08-13",
"open_weights": true,
"license": "Apache-2.0",
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/baichuan-inc/Baichuan-M2-32B"
}
]
}Frequently asked questions
Running locally
Baichuan M2 32B needs an estimated 22.8 GB of VRAM at Q4, an estimated 39.6 GB at Q8 and an estimated 74.1 GB at FP16. All figures are estimates from the 32B parameter count, covering weights plus about 15% and 0.5 GB for a short context.
Yes, it fits on a 24 GB GPU at Q4 with an estimated 22.8 GB. On 16 GB, 12 GB and 8 GB GPUs it does not fit at Q4, Q8 or FP16 (estimate).
Yes. With an estimated 22.8 GB of VRAM at Q4 and the assumption that macOS lets the GPU use about 75% of unified memory, the smallest common Mac memory size that fits Q4 is 32 GB (estimate).
Privacy and data
Yes, Baichuan M2 32B has open weights, so it can be self-hosted for full control of your data.
Capabilities and specs
The context window is 131K tokens (131,072) and the maximum output is 131K tokens (131,072).
No, its only input modality is text.
No, tool calling (function calling) is not supported.
Yes, a reasoning mode is supported.
Quality and benchmarks
The closest alternatives, matched on model kind and release date with no comparable price, are MolmoAct 7B-D by Ai2, Nemotron Nano 9B v2 by NVIDIA and DeepSeek-V3.1 by DeepSeek.
MolmoAct 7B-DBaichuan M2 32B vs MolmoAct 7B-DNemotron Nano 9B v2Baichuan M2 32B vs Nemotron Nano 9B v2DeepSeek-V3.1Baichuan M2 32B vs DeepSeek-V3.1
Licensing and access
It is open weights, released under the Apache-2.0 license.
Yes, the Apache 2.0 license is a permissive license that allows commercial use.
From the Hugging Face repository baichuan-inc/Baichuan-M2-32B.
Basics
Baichuan made Baichuan M2 32B, and it was released on August 13, 2025.
Yes, Baichuan M2 32B is open weights.
Generated from Vertical Tabs data, updated November 2, 2026.
Discussion
No comments yet
Be the first to share your thoughts.