BREAKING
'Build with Gemma' in 20+ Countries
Gemma 4: Open-Weight AI Family
0
Arena AI text
0
%
multilingual MMMLU
0
%
MMMU Pro
Five Variants, Edge to Server
Edge & Laptop
●
E2B and E4B for edge
●
12B laptop-class model
Larger Models
●
26B A4B mixture-of-experts
●
31B dense model
Under the Hood: Gemma 4
1
Thinking Mode
↓
2
Multi-Token Prediction
↓
3
Quantization-Aware Training
↓
4
On-device via LiteRT-LM
Open Community at Gemma's Core
AI NEWS BLITZ
Google is launching Build with Gemma hackathons in more than twenty countries around Gemma 4.