BREAKING
Cerebras Runs Gemma 4 31B at 1,851 t/s
0
t/s
output speed
0
x
vs GPU
0
s
first token
Tokens per Second vs Typical GPU
Cerebras
1851
Typical GPU
50
Vision Agent Loop Now Practical
1
Image input
↓
2
Reasoning
↓
3
Tool calls
↓
4
Verify & retry
Gemma 4 vs Claude Haiku
Gemma 4 31B
Cerebras
●
Intelligence index 29
●
18x faster than Haiku
●
Apache 2.0 license
Claude Haiku
●
Intelligence index 30
●
Comparable quality
First Multimodal Model on Cerebras
AI NEWS BLITZ
Cerebras just launched Google's Gemma 4 31B at over eighteen hundred tokens per second.