Gemma 3 12B
Google · dense · 12B parameters · 131,072 context
Parameters
12B
Context Window
128K tokens
Architecture
Dense
Best GPU
A100 40GB SXM
Cheapest API
$0.10/M
Quality Score
71/100
Intelligence Brief
Gemma 3 12B is a 12B parameter DENSE model from Google, featuring Grouped Query Attention (GQA) with 48 layers and 3,072 hidden dimensions. With a 131,072 token context window, it supports tools, vision, structured output, code, math, multilingual. On standardized benchmarks, it achieves MMLU 74, HumanEval 44, GSM8K 78. The most cost-effective API deployment is via google at $0.10/M output tokens. For self-hosted inference, A100 40GB SXM delivers optimal throughput at $807/month.
Architecture Details
Memory Requirements
BF16 Weights
24.0 GB
FP8 Weights
12.0 GB
INT4 Weights
6.0 GB
GPU Compatibility Matrix
Gemma 3 12B is compatible with 82% of GPU configurations across 41 GPUs at 3 precision levels.
GPU Recommendations
BF16 · 1 GPU · vllm
95/100
score
Throughput
349.9 tok/s
Latency (ITL)
2.9ms
Est. TTFT
0ms
Cost/Month
$807
Cost/M Tokens
$0.88
BF16 · 1 GPU · vllm
95/100
score
Throughput
172.8 tok/s
Latency (ITL)
5.8ms
Est. TTFT
1ms
Cost/Month
$465
Cost/M Tokens
$1.02
BF16 · 1 GPU · vllm
95/100
score
Throughput
156.6 tok/s
Latency (ITL)
6.4ms
Est. TTFT
1ms
Cost/Month
$399
Cost/M Tokens
$0.97
Deployment Options
API Deployment
$0.10/M
output tokens
Single GPU
A100 40GB SXM
$807/mo
Min VRAM: 12 GB
Multi-GPU
RTX 3090 x2
343.7 tok/s
TP· $361/mo
API Pricing Comparison
| Provider | Input $/M | Output $/M | Badges |
|---|---|---|---|
| $0.05 | $0.10 | Cheapest | |
| together | $0.15 | $0.15 |
Cost Analysis
| Provider | Input $/M | Output $/M | ~Monthly Cost |
|---|---|---|---|
| googleBest Value | $0.05 | $0.10 | $1 |
| together | $0.15 | $0.15 | $2 |
Cost per 1,000 Requests
Short (500 tok)
$0.05
via google
Medium (2K tok)
$0.18
via google
Long (8K tok)
$0.60
via google
Performance Estimates
Throughput by GPU
VRAM Breakdown (A100 40GB SXM, BF16)
Precision Impact
bf16
24.0 GB
weights/GPU
~349.9 tok/s
fp8
12.0 GB
weights/GPU
int4
6.0 GB
weights/GPU
Quality Benchmarks
Capabilities
Features
Supported Frameworks
Supported Precisions
Where to Deploy Gemma 3 12B
Similar Models
Gemma 3 4B
4.3B params · dense
Quality: 54
from $0.10/M
Gemma 3 27B
27B params · dense
Quality: 69
from $0.20/M
Amazon Nova Lite
12B params · dense
Quality: 50
from $0.24/M
Mistral Nemo 12B
12B params · dense
Quality: 62
from $0.13/M
Pixtral 12B
12B params · dense
Quality: 50
from $0.15/M
Frequently Asked Questions
How much VRAM does Gemma 3 12B need for inference?
Gemma 3 12B requires approximately 24.0 GB of VRAM at BF16 precision, 12.0 GB at FP8, or 6.0 GB at INT4 quantization. Additional VRAM is needed for KV-cache (393216 bytes per token) and activations (~1.00 GB).
What is the best GPU for Gemma 3 12B?
The top recommended GPU for Gemma 3 12B is the A100 40GB SXM using BF16 precision. It achieves approximately 349.9 tokens/sec at an estimated cost of $807/month ($0.88/M tokens). Score: 95/100.
How much does Gemma 3 12B inference cost?
Gemma 3 12B API inference starts from $0.05/M input tokens and $0.10/M output tokens. Self-hosted inference costs depend on your GPU configuration — use our ROI calculator for a detailed breakdown.