gemma-3-4B-pt Base — benchmark results

Google's pretrained 4B Gemma 3 base checkpoint (March 2025), multimodal with a SigLIP vision encoder and 128K context. Provider: Google. Released 2025-03-12. Access: Open.

Unified ELO 1453 ± 9, rank #981 of 1776 rated models, from 519 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EuroEval Slovak NLU - CSFD Sentiment SK65.92Sentiment classification Score (%)99
EuroEval Czech NLU - CSFD Sentiment68.79Sentiment classification Score (%)98.2
Open Arabic LLM - Aratrust Offensive95.65Accuracy (%)97.2
EuroEval Finnish NLU - Scandisent FI92.64Sentiment classification Score (%)93.8
EuroEval Ukrainian NLU - Cross Domain UK Reviews62.56Sentiment classification Score (%)93.8
EuroEval Swedish NLU - Swerec79.23Sentiment classification Score (%)92
Ukrainian LLM - Long Flores UK RU29.65Score (%)91.7
EuroEval Lithuanian NLU - Atsiliepimai47.92Sentiment classification Score (%)90.8
EuroEval Icelandic NLU - NQII56.87Reading comprehension Score (%)89.8
Open Arabic LLM - Arabic MMLU LAW (Professional)75.8Accuracy (%)89.2
EuroEval Lithuanian NLU - MultiWikiQA LT68.69Reading comprehension Score (%)88.2
IndoBias71.4Ideology and Religion IND (self-reported)88

Interactive version: theaggregate.ai/model?slug=gemma-3-4b-pt-base · How the rankings work · Data refreshed daily, snapshot 2026-07-22.