Gemma 3 12B

Name: Gemma 3 12B
Price: 0.06 USD
Author: Google

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

Gemma 3 12B is a 12.2B-parameter dense model from Google (United States), released 2025-03-12 under the Gemma license. It scores 28.4 on the Small LM Index, #22 of 26 small models. It needs about 9 GB of memory at 4-bit with an 8K-token context and costs $0.060 per million tokens through the cheapest API provider.

Small LM Index

28.4

#22 of 26 scored small models

Likely 22.1–35.8 · from 3 of 6 core benchmarks

Parameters

12.2B

Dense

Price per 1M tokens

$0.060

$0.040 in / $0.12 out · blended 3:1

Context window

128K

tokens

Benchmarks

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

MMLU-Pro Lab-reported Knowledge60.6

Class best: Qwen3.5 9B 82.5

GPQA Diamond Lab-reported Knowledge40.9

Class best: Qwen3.5 9B 81.7

LiveCodeBench Lab-reported Coding · LiveCodeBench v524.6

Class best: Granite 4.2 8B 73.2

Not reported: estimated

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

Humanity's Last Exam≈ 2

AIME≈ 30

SWE-bench Verified≈ 11

Specifications

Lab: Google (US · United States)
Architecture: Dense transformer
Total parameters: 12.2B
Active parameters: All (dense)
Size class: Small
Context window: 128K tokens
Input modalities: text, image
Reasoning: Instruct (no thinking)
License: Gemma
Openness: Open weights
Memory to run: ~9 GB at 4-bit, 8K context (more sizes)
Hugging Face: 359.7K downloads (30 days) · 902 likes
API providers: 2 on OpenRouter
API price: $0.040 in / $0.12 out per 1M (OpenRouter, 2026-11-03)

Weights on Hugging Face ↗Try on OpenRouter ↗

Compare with

Gemma 3 12B vs Ministral 3 14B (14B, index 60.0)
Gemma 3 12B vs Gemma 4 12B (11.9B, index 56.9)
Gemma 3 12B vs Phi-4-reasoning-plus (14.7B, index 54.4)
Gemma 3 12B vs Phi-4 (14.7B, index 44.7)

Sources

https://huggingface.co/google/gemma-3-12b-it

Something wrong here? Report it.

Run it locally

Uses google/gemma-3-12b-it-qat-q4_0-gguf (Q4_0). Swap the tag for another size: Q4_0.

Ollama

ollama run hf.co/google/gemma-3-12b-it-qat-q4_0-gguf:Q4_0

llama.cpp

llama-server -hf google/gemma-3-12b-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-3-12b-it-4bit-DWQ

Text only; for images use mlx-vlm.

vLLM (GPU server)

vllm serve google/gemma-3-12b-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	26 GB	28 GB	34 GB
8-bit Q8_0	14 GB	16 GB	22 GB
4-bit Q4_K_M	8.7 GB	11 GB	17 GB

Check it against your GPU, Mac or phone →