◆ Underrated
Gemma 3n E4B
Google DeepMind🇺🇸 US
multimodal
– API– OpenRouter
Google's on-device multimodal model: handles text, image, video AND audio at just 4B effective parameters. MatFormer architecture enables sub-models within the same checkpoint. Designed for edge deployments.
At a glance
Context
33K tokens
Parameters
8B raw (4B effective)
Input
text, image, video, audio
Specifications
- Status
- Released
- Date
- March 1, 2026
- Lab
- Google DeepMind
- Origin
- US
- Type
- multimodal
- Open weights
- Yes
- License
- Gemma Terms of Use
- Context
- 33K tokens
- Parameters
- 8B raw (4B effective)
- Input
- text, image, video, audio
- Output
- text
Availability
Direct API access–
OpenRouter–
Open weights✓ Yes