BREAKING
Google Expands Gemma 4 Lineup
New 12B Unified Multimodal Model
0
B
parameters
0
layers
0
K
context window
Instruction-Tuned Benchmarks
AIME 2026
77.5
MMLU Pro
77.2
LiveCodeBench
72
MMMU Pro
69.1
QAT vs MTP Tooling
QAT
Q4_0
●
Smaller memory footprint
●
E2B drops to ~1GB
●
Retains more quality
MTP Drafter
~4 layers
●
Speculative decoding
●
~1.4x speedup
●
Ollama and MLX support
Compact Open Models Race Heats Up
AI NEWS BLITZ
Google DeepMind has broadened its open-weight Gemma 4 family with a new unified model.