BREAKING
Cerebras Serves Gemma 4 31B
0
TPS
output speed
0
B
parameters
0
K
context window
Gemma 4 31B Benchmark Scores
MMLU Pro
85.2
AIME 2026
89.2
MMMU Pro
76.9
Gemma 4 31B vs Claude Haiku
Gemma 4 31B
Cerebras
●
1,500+ TPS
●
Index 29
●
Apache 2.0 open weight
Claude Haiku
Proprietary
●
~100 TPS
●
Index 30
●
Closed license
Built for Agentic Loops
1
Real-time UI understanding
↓
2
Document processing
↓
3
Agentic workflows
GA Expected Within the Month
AI NEWS BLITZ
Cerebras is now running Google's Gemma 4 31B at over 1,500 tokens per second.