Gemma 3 4B

Name: Gemma 3 4B
Price: 0.042 USD
Author: Google

Google · Gemma 3 · released 2025-03-12 · data checked 2026-11-01 Needs verification

Gemma 3 4B is a 4.3B-parameter dense model from Google (United States), released 2025-03-12 under the Gemma license. It scores 17.5 on the Small LM Index, #18 of 21 tiny models. It needs about 3.5 GB of memory at 4-bit with an 8K-token context and costs $0.042 per million tokens through the cheapest API provider.

Small LM Index

17.5

#18 of 21 scored tiny models

Likely 13.4–22.7 · from 3 of 6 core benchmarks

Parameters

4.3B

Dense

Price per 1M tokens

$0.042

$0.040 in / $0.050 out · blended 3:1

Context window

128K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best tiny model on that benchmark. Sources: 3 reported by Google. How we label sources.

MMLU-Pro Lab-reported Knowledge43.6

Class best: Qwen3.5 4B 79.1

GPQA Diamond Lab-reported Knowledge30.8

Class best: Nanbeige4.2 3B 87.4

LiveCodeBench Lab-reported Coding · LiveCodeBench v512.6

Class best: LFM3 1.7B 73.5

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 1

AIME≈ 12

SWE-bench Verified≈ 5

Specifications

Lab: Google (US · United States)
Architecture: Dense transformer
Total parameters: 4.3B
Active parameters: All (dense)
Size class: Tiny
Context window: 128K tokens
Input modalities: text, image
Reasoning: Instruct (no thinking)
License: Gemma
Openness: Open weights
Memory to run: ~3.5 GB at 4-bit, 8K context (more sizes)
Hugging Face: 982K downloads (30 days) · 1.6K likes
API providers: 2 on OpenRouter
API price: $0.040 in / $0.050 out per 1M (OpenRouter, 2026-11-03)

Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

Gemma 3 4B vs Nanbeige4.2 3B (4.2B, index 70.4)
Gemma 3 4B vs Gemma 4 E2B (5.1B, index 31.9)
Gemma 3 4B vs Spark-X2.6 4B (4.1B, index 67.5)
Gemma 3 4B vs Spark-X2.5 4B (4.1B, index 65.5)
Gemma 3 4B vs Qwen3.5 4B (4B, index 57.8)
Gemma 3 4B vs Gemma 3n E2B (5.4B, index 14.3)

Sources

https://huggingface.co/google/gemma-3-4b-it

Something wrong here? Report it.

Run it locally

Uses google/gemma-3-4b-it-qat-q4_0-gguf (Q4_0). Swap the tag for another size: Q4_0.

Ollama

ollama run hf.co/google/gemma-3-4b-it-qat-q4_0-gguf:Q4_0

llama.cpp

llama-server -hf google/gemma-3-4b-it-qat-q4_0-gguf:Q4_0

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/gemma-3-4b-it-4bit-DWQ

Text only; for images use mlx-vlm.

vLLM (GPU server)

vllm serve google/gemma-3-4b-it

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	128K context
16-bit F16	9.4 GB	10.0 GB	12 GB
8-bit Q8_0	5.4 GB	5.9 GB	7.9 GB
4-bit Q4_K_M	3.4 GB	4.0 GB	5.9 GB

Check it against your GPU, Mac or phone →