AIbase
EN
Home
AI Models Finder
Gemma 3n E4B Instruct
G

Gemma 3n E4B Instruct

Mobile-first multimodal model from Google using innovative MatFormer (Matryoshka Transformer) architecture, with 8B total/4B effective parameters. Supports text, image, video, and audio input with only 3GB memory footprint, runs efficiently on mobile devices. Supports 140 languages, achieves LMArena score over 1300, first sub-10B model to reach this benchmark.
Intelligence(Relatively Weak)
Speed(Relatively Slow)
Input Supported Modalities
Yes
Is Reasoning Model
32,000
Context Window
32,000
Maximum Output Tokens
2024-06-01
Knowledge Cutoff
Go Compare

Pricing

￥144 /M tokens
Input
￥288 /M tokens
Output
￥0.18 /M tokens
Blended Price

Quick Simple Comparison

Input

Output

Gemini 2.0 Flash Thinking Experimental (Dec '24)
Gemini 1.5 Pro (Sep '24)
￥2.5
Gemini 2.0 Pro Experimental (Feb '25)

Basic Parameters

Gemma 3n E4B InstructTechnical Parameters
Parameter Count
8,000.0M
Context Length
32.00k tokens
Training Data Cutoff
2024-06-01
Open Source Category
Open Weights (Permissive License)
Multimodal Support
Text, Image
Throughput
42
Release Date
2025-06-26
Response Speed
77.03,173 tokens/s

Benchmark Scores

Below is the performance of Gemma 3n E4B Instruct in various standard benchmark tests. These tests evaluate the model's capabilities in different tasks and domains.
Intelligence Index
27.99
Large Language Model Intelligence Level
Coding Index
11.35
Indicator of AI model performance on coding tasks
Math Index
45.37
Capability indicator in solving mathematical problems, mathematical reasoning, or performing math-related tasks
MMLU Pro
48.8
Massive Multitask Multimodal Understanding - Testing understanding of text, images, audio, and video
GPQA
29.6
Graduate Physics Questions Assessment - Testing advanced physics knowledge with diamond science-level questions
HLE
4.4
The model's comprehensive average score on the Hugging Face Open LLM Leaderboard
LiveCodeBench
14.6
Specific evaluation focused on assessing large language models' ability in real-world code writing and solving programming competition problems
SciCode
8.1
The model's capability in code generation for scientific computing or specific scientific domains
HumanEval
75
Score achieved by the AI model on the specific HumanEval benchmark test set
Math 500 Score
77.1
Score on the first 500 larger, more well-known mathematical benchmark tests
AIME Score
13.7
An indicator measuring an AI model's ability to solve high-difficulty mathematical competition problems (specifically AIME level)
AIbase
Empowering the Future, Your AI Solution Knowledge Base
English简体中文繁體中文にほんご
© 2025AIbase