LingSante-3.1-flash-MXFP4
a quantized version of the giant medical MoE model
2 min read
The new open model from inclusionAI combines complex medical reasoning, professional healthcare communication, evidence-based research, and long-term tasks. The architecture is built on LingSante-3.1-flash using a hybrid-linear Mixture-of-Experts approach (MoE — a mixture of experts where only a subset of parameters is activated for each token). The model contains around 560 billion parameters in total (with 25 billion active per token) and supports a context window of up to 262,144 tokens. This version uses MXFP4 quantization (a rounding-to-nearest method without additional training), which significantly lowers memory requirements while retaining core capabilities.
Key features
- Complex medical reasoning: identifying key clinical decision factors and checking the evidence base.
- Long-term medical tasks: coordinating tools and information within a single dynamic scenario.
- Professional behavior: adapting responses to real clinical scenarios and safety standards.
- MXFP4 mixed quantization: optimizing expert module weights for efficient deployment.
What you can do
- Medical research: use for literature analysis, evidence synthesis, and clinical document processing.
- Agent development: create specialized assistants for working with medical data and search.
- Information extraction: build structured data extraction systems for healthcare.
Example request to a deployed server via SGLang API with reasoning support:
curl -N -sS "http://localhost:30080/v1/chat/completions" \
-H 'Content-Type: application/json' \
-d '{
"model": "LingSante-3.1-flash-MXFP4",
"messages": [{"role": "user", "content": "What is the capital of France?"}],
"chat_template_kwargs": {"enable_thinking": true},
"stream": true,
"max_tokens": 512
}'Read more
DetailsBase model: LingSante-3.1-flash
Checkpoint size: 291.72 GiB
Recommended hardware: 4 × NVIDIA H20-3e (TP4 / EP4 / DP1)
Limitations: not a medical device and does not replace qualified professionals










Comments