Biomedical · Google

MedGemma 4B

Hugging Face: google/medgemma-4b-it · checked 2026-10-13

The original 4B multimodal MedGemma.

Datasheet

Maker	Google
Parameters	4.30 BVerifiedsafetensors.total from the HF API Source
Licence	Health AI Developer Foundations termsVerified
Access	Gated: accept the terms on Hugging Face to download
Task (HF pipeline)	image-text-to-text
First published on HF	2025-05-19
Last updated on HF	2025-10-28
Weights in memory	8.6 GB at 16-bit · 2.6 GB at ~4.85 bitsEstimateParameters × bits ÷ 8. Excludes context cache and runtime overhead; not every model has a 4-bit build.

Benchmark

64.4%

MedQA (4-option), accuracy · #3 of 3 scored in this category

"MedGemma 1 4B" column, text evaluations table. Source

Other published results

MedMCQA: 55.7%Sourced Source

See the full biomedical ranking

The description summarises the model card at https://huggingface.co/google/medgemma-4b-it. Parameters and licence are read from the Hugging Face API; the weights-in-memory line is our arithmetic, not a measurement.