gemma-4-E2B-it: benchmark results

Instruction-tuned effective-2B tier of Google's Apache-2.0 Gemma 4 on-device family (April 2026) with native vision and audio input. Provider: Google. Released 2026-04-02. Access: Open.

Unified ELO 1492 ± 1, rank #727 of 1392 rated models, from 375 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EuroEval Italian Summarization - Ilpost SUM38.97Score (%)98.7
Enkrypt AI - Insecure Code Risk4Risk Score94.5
EuroEval Albanian Summarization - LR SUM SQ33.87Score (%)92.5
EuroEval Danish Summarization - Nordjylland News37.24Score (%)89.9
EuroEval Portuguese NLU - ScaLA PT36.48Linguistic acceptability Score (%)88.2
EuroEval Italian NLU - ScaLA IT40.1Linguistic acceptability Score (%)86.4
EuroEval Croatian NLU - Multi Wiki QA HR65.47Reading comprehension Score (%)85.2
MERA - ruHumanEval47.32pass@1 (%)84.6
EuroEval Spanish Summarization - Mlsum ES28.48Score (%)84.4
EuroEval French NLU - ScaLA FR45.98Linguistic acceptability Score (%)84.3
EuroEval Swedish Summarization - Swedn37.56Score (%)83.2
EuroEval Icelandic Summarization - RRN37.23Score (%)81.8

Interactive version: theaggregate.ai/model?slug=gemma-4-e2b-it · How It Works · Data refreshed daily, snapshot 2026-09-05.