← All models

Gemma 3

Google · 2025 · Dense multimodal

Gemini-derived open models from 270M to 27B. Vision, 140+ languages, and up to 128K context: a full ladder from phones to desktops, all local in Delta.

Maker
Google
Architecture
Dense multimodal
Fits from
Phone / edge · 3 GB+ RAM
Context
up to 33K
Runs
Local Intelligence
VisionMultilingualGeneral chatMultilingual workVision + text

Sizes

Gemma 3 270MQ8_0

requires 3 GB+ memory · 33K context · Phone / edge

205 MBWeights
Gemma 3 1BQ8_0

requires 4 GB+ memory · 8K context · Phone / edge

717 MBWeights
Gemma 3 4BQ4_K_M

requires 6 GB+ memory · 33K context · Phone / edge

2.5 GBWeights
Gemma 3 12BQ4_K_M

requires 13 GB+ memory · 33K context · Desktop

7.3 GBWeights
Gemma 3 27BQ4_K_M

requires 27 GB+ memory · 33K context · Workstation

17 GBWeights

Memory at context length

Approximate working set for Gemma 3 270M as context grows.

4K
307 MB
8K
307 MB
16K
410 MB
33K
410 MB