Gemma 4 31B
Non-reasoning
Creator
Google DeepMind
Released
2026-04-02
Intelligence
32.3
Artificial Analysis Index
Coding
33.9
Artificial Analysis Index
In $/1M
$0.14
input tokens
Out $/1M
$0.40
output tokens
Blended $/1M
$0.20
3:1 blended
Speed
61
tokens / sec
Benchmark breakdown
Independent evaluation scores, normalized to 0–100.
GPQA Diamond
76
Humanity’s Last Exam
12
MMLU-Pro—
SciCode
41
LiveCodeBench—
MATH-500—
AIME 2025—
τ²-Bench (agentic)
65
Terminal-Bench Hard
30
IFBench
53