Gemma 4 12B
Parameters
11.95B
Context Length
262K
Modality
Multimodal
Auxiliary Parameters
550M
Architecture
Dense
License
Apache-2.0
Release Date
3 Jun 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Input: $0.10 · Output: $0.30
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
2x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
262,144 tokens
Consumer
6x RTX 4090
24GB VRAM
Datacenter
2x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
Rank
#109
| Benchmark | Score | Rank |
|---|---|---|
Professional Knowledge | 0.772 | 32 |
Graduate-Level QA | 0.788 | 68 |
Coding Index | auto 0.31 | 124 |
Intelligence Index | auto 0.14 Standard 0.09 | 213 256 |
Rankings
Overall Rank
#109
Coding Rank
#116
About Gemma 4 12B
Google DeepMind's 12B dense open-weights model released June 3, 2026, bridging the gap between the edge-friendly E4B and the more advanced 26B MoE. Uniquely features an encoder-free unified architecture that projects raw image patches and audio waveforms directly into the LLM embedding space through lightweight linear layers, eliminating the latency and memory overhead of separate encoders. Supports 256K token context, native text/image/audio inputs, configurable thinking mode, and runs on consumer laptops with 16GB of RAM.
Technical Specifications
Attention
Attention Structure
Multi-Head Attention
Attention Heads
16
Key-Value Heads
8
Attention Head Dimension
256
Position Embedding
Absolute Position Embedding
RoPE Theta
10,000
Sliding Window Attention
Yes
Sliding Window Size
1,024
Sliding Window Ratio
83.3%
Linear Attention
No
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
GELU
Dimensions
Auxiliary Parameters
550M
Hidden Dimension Size
3,840
Number of Layers
48
FFN Intermediate Size (Dense)
15,360
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
262,144
About Gemma 4
Gemma 4 is Google DeepMind's most advanced open model family, built from Gemini 3 research and technology. Featuring both Dense and Mixture-of-Experts (MoE) architectures, these multimodal models handle text, images, and audio (on smaller variants), with context windows up to 256K tokens. Designed for frontier-level performance across reasoning, coding, and agentic workflows, Gemma 4 delivers unprecedented intelligence-per-parameter from mobile devices to enterprise servers. Released under Apache 2.0 license.