Vision-language · Google

Gemma 3 12B

Hugging Face: google/gemma-3-12b-it · checked 2026-10-13

Multimodal Gemma with image input and 128K context.

Datasheet

Maker	Google
Parameters	12.19 BVerifiedsafetensors.total from the HF API Source
Licence	Gemma Terms of UseVerified
Access	Gated: accept the terms on Hugging Face to download
Task (HF pipeline)	image-text-to-text
First published on HF	2025-03-01
Last updated on HF	2025-03-21
Weights in memory	24 GB at 16-bit · 7.4 GB at ~4.85 bitsEstimateParameters × bits ÷ 8. Excludes context cache and runtime overhead; not every model has a 4-bit build.

Benchmark

Not yet measured

MMMU (val)

No comparable public score for this benchmark yet (Model card is gated; not read). We will not estimate one.

Other published results

BFCL V4 overall (prompt mode): 30.43%Sourced Source
LMArena text, non-English: 1,317.9Sourced Source

See the full vision-language ranking

The description summarises the model card at https://huggingface.co/google/gemma-3-12b-it. Parameters and licence are read from the Hugging Face API; the weights-in-memory line is our arithmetic, not a measurement.