← All models

Google2026

Gemma 4

Multimodal · audio + vision · edge

Google’s most capable open models for mobile and IoT. Audio and vision for real-time edge processing, with near-zero latency on phones, Raspberry Pi, and Jetson-class boards, offline, on your silicon.

  • Fits fromLaptop · 7 GB+
  • Contextup to 33K
  • Sizes2
  • RunsLocal on Delta
VisionAudioEdgeMultilingualPhones & IoTRealtime visionAudio + text

Sizes

Gemma 4 E2B

3.1 GB • 33k ctx • Q4_K_M

Gemma 4 E4B

5 GB • 33k ctx • Q4_K_M

Memory at context length

Approximate working set for Gemma 4 E2B as context grows.

4K
3.9 GB
8K
4.1 GB
16K
4.4 GB
33K
4.6 GB