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.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| gemma-3-270m | 268M | 0.6 GB | 0.3 GB | 0.1 GB | RTX 3070 | $0.088/hr | 0.10 GB |
| gemma-3-1b-it | 1000M | 2.2 GB | 1.1 GB | 0.6 GB | RTX 3070 | $0.088/hr | 0.14 GB |
| gemma-3-4b-it | 4.3B | 9.6 GB | 4.8 GB | 2.4 GB | RTX 3060 | $0.110/hr | 0.74 GB |
| gemma-3-12b-it | 12.2B | 27.2 GB | 13.6 GB | 6.8 GB | RTX A6000 | $0.363/hr | 2.3 GB |
| gemma-3-27b-it | 27.4B | 61.3 GB | 30.7 GB | 15.3 GB | A100 | $1.31/hr | 2.9 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)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| gemma-3-270m | 268M | BF16 | 0.6 GB | RTX 3070 | $0.088/hr |
| gemma-3-270m-it | 268M | BF16 | 0.6 GB | RTX 3070 | $0.088/hr |
| gemma-3-1b-it | 1000M | BF16 | 2.2 GB | RTX 3070 | $0.088/hr |
| gemma-3-1b-pt | 1000M | BF16 | 2.2 GB | RTX 3070 | $0.088/hr |
| gemma-3-4b-it | 4.3B | BF16 | 9.6 GB | RTX 3060 | $0.110/hr |
| gemma-3-12b-it | 12.2B | BF16 | 27.2 GB | RTX A6000 | $0.363/hr |
| gemma-3-27b-it | 27.4B | BF16 | 61.3 GB | A100 | $1.31/hr |
Quantized versions (1)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| gemma-3-27b-it-FP8-dynamic | 27.4B | F8_E4M3 | 30.7 GB | RTX 5000 Ada | $0.539/hr |
Fine-tunes and community models (3)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| gemma-3-4B-T1-it | 3.9B | BF16 | 8.7 GB | RTX 3060 | $0.110/hr |
| gemma-3-12b-it-heretic | 12.2B | BF16 | 27.2 GB | RTX A6000 | $0.363/hr |
| gemma-3-12b-it-heretic-v2 | 12.2B | BF16 | 27.2 GB | RTX 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
- How to run Gemma 3: engine support, chat template and context flags, with a serving command sized per model.
- How to fine-tune Gemma 3: LoRA and QLoRA memory per model, and the license limits on derivatives.