gemma-3-12B-pt: benchmark results

Google's Gemma 3 12B pretrained base model - open weights with text+image input and a 128K context (March 2025). Provider: Google. Released 2025-03-12. Access: Open.

Unified ELO 1549 ± 1, rank #439 of 1392 rated models, from 526 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EuroEval Czech NLU - CSFD Sentiment75.96Sentiment classification Score (%)100
EuroEval Slovak NLU - CSFD Sentiment SK68.04Sentiment classification Score (%)100
EuroEval Swedish NLU - Swerec81.27Sentiment classification Score (%)100
EuroEval Ukrainian NLU - Cross Domain UK Reviews65.19Sentiment classification Score (%)100
LatamBoard - Spanish COPA85.8Score (%)100
LatamBoard - Spanish MGSM30.8Score (%)100
Ukrainian LLM - ARC Easy UK80.64Score (%)100
Ukrainian LLM - Global MMLU Full UK College Biology86.81Score (%)100
Ukrainian LLM - Global MMLU Full UK High School Geography85.86Score (%)100
Ukrainian LLM - Global MMLU Full UK High School Government AND Politics86.53Score (%)100
Ukrainian LLM - Global MMLU Full UK High School Psychology88.26Score (%)100
Ukrainian LLM - Global MMLU Full UK Security Studies79.59Score (%)100

Interactive version: theaggregate.ai/model?slug=gemma-3-12b-pt · How It Works · Data refreshed daily, snapshot 2026-09-05.