Ministral 14B
Trinity Nano Preview
12B
- Context
- 33Ktokens
- Released
- 1y ago
- Size
- 5.4Bparameters
Efficient on-device Gemma 3n model for text, image, and audio understanding
- Multimodal model
- Open weights · 5.4B
- Accepts images, audio
How it compares
Size class
Small5.4BAn open-weight model of 5 to 25 billion parameters, the size that runs on a single consumer GPU.
Open models of a similar size from other labs
Vs. your stack
Add the models you use to see how this one compares.
Sign inFamily tree
Models trained from one another, as their makers describe it.
Gemma 3n E4B ITGoogle DeepMind · 2025
Additional source evidence
- Source model
- google/gemma-3n-e2b-it
- Source family
- gemma
- Last updated
- 2025-06-26
- Knowledge cutoff
- 2024-06
- Model context
- 33K
- Model output limit
- 33K
- Open weights
- Yes
- File attachments
- Yes
- Reasoning
- No
- Tool calling
- No
- Temperature control
- Yes
Downloadable weights
Complete source record (JSON)
{
"id": "google/gemma-3n-e2b-it",
"name": "Gemma 3n E2B IT",
"description": "Efficient on-device Gemma 3n model for text, image, and audio understanding",
"family": "gemma",
"last_updated": "2025-06-26",
"knowledge": "2024-06",
"attachment": true,
"reasoning": false,
"tool_call": false,
"temperature": true,
"modalities": {
"input": [
"text",
"image",
"audio"
],
"output": [
"text"
]
},
"limit": {
"context": 32768,
"output": 32768
},
"release_date": "2025-06-26",
"open_weights": true,
"license": "Gemma Terms of Use",
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/google/gemma-3n-E2B-it"
}
]
}Text provenance
Gemini app/webGoogle SynthID text watermarking
Google DeepMind says text generated in the Gemini app and web experience is watermarked with SynthID by adjusting token probabilities.
Official explanationFrequently asked questions
Running locally
About 4.3 GB at Q4, 7.1 GB at Q8 and 13 GB at FP16, based on the 5.4B parameter count. All figures are estimates: weights plus about 15% and 0.5 GB for a short context.
Yes. A 24 GB and a 16 GB GPU fit at FP16 (estimated 13 GB), and a 12 GB and an 8 GB GPU fit at Q8 (estimated 7.1 GB).
Yes, at Q4 the estimated VRAM is 4.3 GB. Assuming macOS lets the GPU use about 75% of unified memory, the smallest common Mac memory size that fits Q4 is 8 GB (estimate).
Privacy and data
Yes, Gemma 3n E2B IT has open weights, so you can self-host it for full control of your data.
Capabilities and specs
The context window is 33K tokens (32,768), and the maximum output is 33K tokens (32,768).
Images and audio: yes. Video and PDFs: no. The input modalities are text, image and audio.
No, tool calling (function calling) is not supported.
No, a reasoning or thinking mode is not supported.
The knowledge cutoff is 2024-06.
Quality and benchmarks
ERNIE 4.5 0.3B and ERNIE 4.5 VL 28B A3B by Baidu, and DiffuCoder 7B by Apple. They were matched on model kind and release date, with no comparable price.
ERNIE 4.5 0.3BGemma 3n E2B IT vs ERNIE 4.5 0.3BERNIE 4.5 VL 28B A3BGemma 3n E2B IT vs ERNIE 4.5 VL 28B A3BDiffuCoder 7BGemma 3n E2B IT vs DiffuCoder 7B
Versions and lineage
Both have a context window of 33K tokens (32,768) and both were released on June 26, 2025, so on these facts they do not differ.
Gemma 3n E2B IT is fine-tuned from Gemma 3n E4B IT.
Licensing and access
It is open weights, licensed under the Gemma Terms of Use.
From the Hugging Face repository google/gemma-3n-E2B-it.
Basics
Google DeepMind released Gemma 3n E2B IT on June 26, 2025.
Yes, Gemma 3n E2B IT is available as open weights.
Generated from Vertical Tabs data, updated November 10, 2026.
Discussion
No comments yet
Be the first to share your thoughts.