BREAKING
Gemma 4 31B Hits 1,851 Tokens/sec
0tok/s
Throughput
0x
vs typical GPU
0s
Time to first token
Open Speech-to-Speech Stack
1Parakeet ASR
2Gemma 4 on Cerebras
3Text-to-speech
Gemma 4 31B Benchmarks
MMLU Pro85.2
AIME 202689.2
LiveCodeBench80
MMMU Pro76.9
Strengths and Caveats
UpsidesPraise
Fluid voice demo
Stable JSON output
Apache 2.0 license
CaveatsFlagged
Reasoning stalls on hard prompts
Heavy local execution
Quantized GGUF builds help
Open Voice Agents Go Real-Time
AI NEWS BLITZ
Google's open Gemma 4 31B now powers near-instant voice AI on Cerebras hardware.