Gemma 4 31B (IT) (Non-reasoning): benchmark results

Provider: Google. Released 2026-04-02. Access: Open.

Unified ELO 1572 ± 1, rank #821 of 3078 rated models, from 30 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
CPTU Bench4.32Average Score (1-5)96
FrameBench - Japanese97.6Accuracy (%; mean of five prompt templates)92
MERA v2 - Enantiosemy55.8Score (%)81.6
MERA v2 - Humor36.4Score (%)81.6
MERA v2 - Reasoning62.4Score (%)81.6
MERA v2 - NewReasoning68.9Score (%)78.9
MERA v2 - SAGE68Score (%)78.9
FrameBench - Frame Identification - English81Accuracy (%; FrameNet candidate frames)76.3
MERA v2 - Culture Specific35.9Score (%)76.3
FrameBench - English97.5Accuracy (%; mean of five prompt templates)73.7
MERA v2 - Characters57.5Score (%)73.7
MERA v2 - RUBIN63.7Score (%)73.7

Interactive version: theaggregate.ai/model?slug=gemma-4-31b-it-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-19.