Gemma 3 27B: benchmark results

Google's open-weight Gemma 3 27B, the flagship multimodal model of the Gemma 3 family (March 2025). Provider: Google. Released 2025-03-12. Access: Open.

Unified ELO 1513 ± 1, rank #631 of 1392 rated models, from 616 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
OpenVLM LLaVA-Bench - Complex112.4Score100
OpenVLM LLaVA-Bench - Detail102.9Score99.6
OpenVLM MMT-Bench - Chemical Apparatus Recognition80Score (%)99
OpenVLM MMMU - Energy and Power73.3Accuracy (%)98.8
LLaVA-Bench103.1Score98.7
OpenVLM MMT-Bench - Rock Recognition80Score (%)98.3
OpenVLM MMMU - Electronics86.7Accuracy (%)98.2
OpenVLM MMT-Bench - Art Design90.9Score (%)98.1
OpenVLM MMBench V1.1 CN - Identity Reasoning100Accuracy (%)97.9
OpenVLM MMMU - History83.3Accuracy (%)96.5
OpenVLM MMT-Bench - DU86.4Score (%)96.4
OpenVLM MMMU - Art Theory93.3Accuracy (%)96.1

Interactive version: theaggregate.ai/model?slug=gemma-3-27b · How It Works · Data refreshed daily, snapshot 2026-09-05.