BREAKING
Google Unveils Gemma 4 12B Unified
Encoder-Free Architecture
1
Raw image & audio
↓
2
Lightweight embedding
↓
3
LLM space
0
B
Parameters
0
K
Context tokens
0
GB
Q4_0 memory
Benchmark Scores
GPQA Diamond
78.8
LiveCodeBench
72
DocVQA
94.9
Runs Locally on Laptops
On-device use
●
Coding agents
●
Offline image & video
●
Speech transcription
Caveats
●
Settings affect reasoning
●
Trails 35B-class models
●
GGUF still maturing
Open Weights, Apache 2.0
AI NEWS BLITZ
Google DeepMind has released Gemma 4 12B, an encoder-free open-weight multimodal model.