Gemma 3 models

11 Gemma 3 models from Google on Hugging Face, from 268M to 27.4B parameters, published by Google in BF16, with contexts from 32K to 128K tokens. The smallest official model, gemma-3-270m, needs about 0.6 GB of VRAM at its published precision; the cheapest live fit is RTX 3070 at $0.088/hr.

Part of the Gemma series · Previous generation: Gemma 2 · Next generation: Gemma 3n

Pick a size

One row per official Gemma 3 size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.

ModelParametersNative VRAMFP8 VRAMINT4 VRAMLive GPU fit (native)Est. $/hrKV cache at 32K
gemma-3-270m268M0.6 GB0.3 GB0.1 GBRTX 3070$0.088/hr0.10 GB
gemma-3-1b-it1000M2.2 GB1.1 GB0.6 GBRTX 3070$0.088/hr0.14 GB
gemma-3-4b-it4.3B9.6 GB4.8 GB2.4 GBRTX 3060$0.110/hr0.74 GB
gemma-3-12b-it12.2B27.2 GB13.6 GB6.8 GBRTX A6000$0.363/hr2.3 GB
gemma-3-27b-it27.4B61.3 GB30.7 GB15.3 GBA100$1.31/hr2.9 GB

VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.

Official models (7)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
gemma-3-270m268MBF160.6 GBRTX 3070$0.088/hr
gemma-3-270m-it268MBF160.6 GBRTX 3070$0.088/hr
gemma-3-1b-it1000MBF162.2 GBRTX 3070$0.088/hr
gemma-3-1b-pt1000MBF162.2 GBRTX 3070$0.088/hr
gemma-3-4b-it4.3BBF169.6 GBRTX 3060$0.110/hr
gemma-3-12b-it12.2BBF1627.2 GBRTX A6000$0.363/hr
gemma-3-27b-it27.4BBF1661.3 GBA100$1.31/hr

Quantized versions (1)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
gemma-3-27b-it-FP8-dynamic27.4BF8_E4M330.7 GBRTX 5000 Ada$0.539/hr

Fine-tunes and community models (3)

ModelParametersPublished asVRAM neededLive GPU fitEst. $/hr
gemma-3-4B-T1-it3.9BBF168.7 GBRTX 3060$0.110/hr
gemma-3-12b-it-heretic12.2BBF1627.2 GBRTX A6000$0.363/hr
gemma-3-12b-it-heretic-v212.2BBF1627.2 GBRTX A6000$0.363/hr

VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.

Run and fine-tune Gemma 3

More from Google

Start ComfyUI on a rented GPU.

© 2026 Aquanode. All rights reserved.

All trademarks, logos and brand names are the property of their respective owners.