BREAKING
Google Publishes Gemma 4 Report
Gemma 4 Model Lineup
E2B / E4B
edge
●
Effective 2.3B / 4.5B
●
Phone and edge devices
12B / 31B
desktop
●
12B is encoder-free
●
31B for consumer GPUs
0
M
old encoder
0
M
new projection
0
K
context
0
%
KV cache cut
0
GB
31B bf16
0
GB
31B quantized
Arena AI Text Scores
Gemma 4 31B
1451
26B-A4B
1438
Gemma 3 27B
1365
Over 150M Gemma Downloads
AI NEWS BLITZ
Google DeepMind has released the technical report for its open-weight Gemma 4 models.