Model

Gemma 3 12B

OPEN

Google · Gemma 3 family · released Mar 12, 2025

Sweet spot for 12GB consumer cards; dense decoder + shared SigLIP vision encoder, 128K context.

ReasoningCodingVisionFunction callingTool useAgentic

944.3

Elo · rank #224

Parameters

12.2B

Active params

12.2B (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

AI2DVision84.2%#29

best: Molmo 72B · 96.3%

BIG-Bench HardReasoning85.7%#15

best: ERNIE 4.5 300B-A47B · 94.3%

BIRD-SQLCoding47.9%#6

best: Command A · 59.5%

ChartQAVision75.7%#36

best: MiniMax-VL-01 · 91.7%

DocVQAVision87.1%#39

best: Qwen2-VL-72B · 96.5%

FACTS GroundingKnowledge75.8%#7

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

GPQA DiamondReasoning40.9%#214

best: GPT-5.6 · 94.6%

GSM8KMath94.4%#16

best: Llama 3.1 405B · 96.8%

HumanEvalCoding85.4%#41

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning88.9%#30

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding24.6%#160

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

MATH-500Math83.8%#60

best: GPT-5 · 99.4%

MathVistaVision62.9%#38

best: Seed 2.1 Pro · 90.7%

MBPPCoding73.0%#30

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

MMLU-ProKnowledge60.6%#113

best: Claude Fable 5 · 91.5%

MMLUKnowledge69.5%#147

best: OpenAI o3 · 92.9%

MMMUVision59.6%#78

best: Claude Fable 5 · 89.3%

SimpleQAKnowledge6.3%#44

best: GPT-4.5 · 62.5%

TextVQAVision67.7%#23

best: Molmo 2 8B · 85.7%

VQAv2Vision71.6%#16

best: Reka Edge (2026) · 88.4%

Run it locally

VRAM @ Q4

6.6 GB

VRAM @ FP16

24 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-3-12b-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)