Project AURIONAI Use-Cases Research & Innovation
edge LLM

Gemma 3n E4B

Gemma 3n E4B on-device multimodal model for phones, laptops and tablets.

Google DeepMindGoogle DeepMindUK / USAGemma 3nFamily
Google DeepMind
Levelmodel
Classedge LLM
Parametersnot in the notes
Context32,000
Released3 Jun 2025
Opennessopen weights restricted
Runs onmobile or NPU, laptop, offline capable
Statusactive

Benchmark Results

vendor-reported
ARC-cknowledge and reasoning
PT (pretrained checkpoint)vendor-reported
61.6
050100

Score in %. Higher is better.

ai.google.dev
Global MMLUknowledge and reasoning
ITvendor-reported
60.3
050100

Score in %. Higher is better.

ai.google.dev
GPQA Diamondknowledge and reasoning
ITvendor-reported
23.7
050100

Score in %. Higher is better.

ai.google.dev
HellaSwagknowledge and reasoning
PT (pretrained checkpoint)vendor-reported
78.6
050100

Score in %. Higher is better.

ai.google.dev
LiveCodeBench v5 v5coding
ITvendor-reported
25.7
050100

Score in %. Higher is better.

ai.google.dev
MGSMmath
ITvendor-reported
60.7
050100

Score in %. Higher is better.

ai.google.dev
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).

Lineage

upcycle prune expand
Gemma 3n E4B to Gemma 3n E2B: upcycle prune expand, medium confidenceGemma 3n E4BGemma 3n E4BGoogle DeepMind, Jun 2025Gemma 3n E2BGemma 3n E2BGoogle DeepMind, Jun 2025
Google DeepMindGemma 3n E2B is a child (upcycle prune expand)medium confidenceHugging Face base_model
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 2 models. Confidence and evidence are the notes' own.

R&D Topics

Quantization and edge inferenceproduced bySmall language models (SLM)produced byVisual instruction tuningproduced by

Identity

Developer
Google DeepMind
Family
Gemma 3n
Released
3 Jun 2025 sourcesregistry date
Weights published
–
Record checked
10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
other MatFormer; MobileNet-V5 vision encoder; audio input at 6.25 tokens/s
Total parameters
– sources
Active parameters
–
Checkpoint parameters
7.85B
Note
E4B-it raw parameter total on HF. Published on HF 3 Jun 2025 (E4B-it).
Context length
32,000 tokens
Input
text, image, audio
Output
text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
open weights restricted
Licence
Gemma Terms of Use sources
Access routes
open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
mobile or NPU, laptop, offline capable
Footprint
Footprint in GB and measured tok/s on a named device unverified
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
not stated
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
Trained on over 140 languages
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Industry Use Cases

No industry use case links this model.

Sources

  1. Descripción general del modelo Gemma 3n  |  Google AI for Developers ai.google.dev https://ai.google.dev/gemma/docs/gemma-3n
  2. Gemma 3n model card  |  Google AI for Developers ai.google.dev https://ai.google.dev/gemma/docs/gemma-3n/model_card
  3. huggingface.co/api/models/google/gemma-3n-E4B-it https://huggingface.co/api/models/google/gemma-3n-E4B-it
  4. Gemma 3n , Google DeepMind deepmind.google https://deepmind.google/models/gemma/gemma-3n/

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Gemma 3n E4B", https://aurion.semai.id/models/google-gemma-3n-e4b/, records of October 2026.