Gemma 4 31B (High): benchmark results

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

Unified ELO 1678 ± 1, rank #205 of 3078 rated models, from 18 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
ObviousBench100Answer pass³ (%)97.7
Korean CSAT 2026 (Easy Mode) - English100Points (out of 100)90.5
Korean CSAT 2026 (Easy Mode) - Mathematics98.67Points (out of 100)71.4
PetriBench - Boundedness (Hard)52Exact-Match Accuracy (%)71.4
Korean CSAT 2026 (Easy Mode) - Korean97Points (out of 100)70.4
Korean CSAT 2026 (Easy Mode) - Total419.67Points (out of 450)65.6
Korean CSAT 2026 (Easy Mode) - Life Science I40Points (out of 50)62.9
Korean CSAT 2026 (Easy Mode) - Society and Culture40Points (out of 50)59.2
Korean CSAT 2026 (Easy Mode) - Chemistry I43Points (out of 50)53.7
PetriBench - L4-Liveness (Hard)54Exact-Match Accuracy (%)51.4
PetriBench - Deadlock (Hard)51Exact-Match Accuracy (%)42.9
PetriBench - Hard25.8Exact-Match Accuracy (%)37.1

Interactive version: theaggregate.ai/model?slug=gemma-4-31b-high · How It Works · Data refreshed daily, snapshot 2026-09-20.