BREAKING
Google Releases Encoder-Free Gemma 4 12B
How encoder-free works
1
Raw input
↓
2
Linear projection
↓
3
Single decoder Transformer
0
M
image embedder
0
M
old vision encoder
0
M
removed audio encoder
Quantized to fit 16GB laptops
FP16
26.7
Quantized
6.7
0
GPQA Diamond
0
LiveCode Bench
0
tokens/sec local
Open weights, Apache 2.0
AI NEWS BLITZ
Google DeepMind has launched Gemma 4 12B, a unified multimodal model with no dedicated encoders.