Model

Gemma 3 4B

OPEN

Google · Gemma 3 family · released Mar 12, 2025

Compact multimodal Gemma 3 for edge deployment; dense decoder + shared SigLIP vision encoder, 128K context.

ReasoningCodingVisionFunction callingTool useAgentic

680.3

Elo · rank #269

Parameters

4.3B

Active params

4.3B (dense)

Context

128K tokens

Architecture

Dense decoder-only Transformer, GQA + 5:1 local:global attention, shared frozen SigLIP-400M vision encoder

License

Gemma

Languages

140+

API price (in/out)

No hosted API

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AI2DVision74.8%#38

best: Molmo 72B · 96.3%

BIG-Bench HardReasoning72.2%#47

best: ERNIE 4.5 300B-A47B · 94.3%

BIRD-SQLCoding36.3%#9

best: Command A · 59.5%

ChartQAVision68.8%#38

best: MiniMax-VL-01 · 91.7%

DocVQAVision75.8%#41

best: Qwen2-VL-72B · 96.5%

FACTS GroundingKnowledge70.1%#10

best: Gemini 2.5 Flash-Lite (Thinking) · 86.8%

GPQA DiamondReasoning30.8%#257

best: GPT-5.6 · 94.6%

GSM8KMath89.2%#46

best: Llama 3.1 405B · 96.8%

HumanEvalCoding71.3%#82

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning90.2%#16

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding12.6%#167

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math75.6%#79

best: GPT-5 · 99.4%

MathVistaVision50.0%#54

best: Seed 2.1 Pro · 90.7%

MBPPCoding63.2%#51

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLU-ProKnowledge43.6%#146

best: Claude Fable 5 · 91.5%

MMLUKnowledge58.1%#212

best: OpenAI o3 · 92.9%

MMMUVision48.8%#98

best: Claude Fable 5 · 89.3%

SimpleQAKnowledge4.0%#46

best: GPT-4.5 · 62.5%

TextVQAVision57.8%#27

best: Molmo 2 8B · 85.7%

VQAv2Vision62.4%#19

best: Reka Edge (2026) · 88.4%

Run it locally

VRAM @ Q4

2.6 GB

VRAM @ FP16

8 GB

Fits on (Q4)

RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

GGUF Q4_K_M · GGUF Q8_0 · AWQ · MLX · QAT Int4

Gemma 3 family

Elo progression across releases

Gemma 3 12BMar 2025944.3 Gemma 3 27BMar 20251001.5 Gemma 3 4BMar 2025680.3

Resources

HFgoogle/gemma-3-4b-it GHgoogle-deepmind/gemma DOCSai.google.dev/gemma/docs/core APIAvailable via Google API

API price weights · each benchmark row carries its own source badge (see methodology)