Gemma 4 26B A4B (Reasoning) — benchmark results

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

Unified ELO 1573 ± 21, rank #612 of 1839 rated models, from 35 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA IFBench72.45Accuracy (%)89.3
AA Humanity's Last Exam18.26Accuracy (%)76.5
Epoch AI - Scicode40.05Score72.7
AA GPQA Diamond79.19Accuracy (%)70.8
Artificial Analysis Intelligence Index25.69Intelligence Index68.6
AA Long Context Reasoning55.67Accuracy (%)64.6
AA Omniscience - Science, Engineering & Mathematics30.1Accuracy (%)64
AA Omniscience - Business18.6Accuracy (%)58.6
AA MMMU-Pro69.25Accuracy (%)55.7
AA Omniscience - Software Engineering (SWE) - Swift40Accuracy (%)55.2
Tau3 Banking11.75Success Rate (%)53.6
AA Terminal-Bench Hard13.64Accuracy (%)50.1

Interactive version: theaggregate.ai/model?slug=gemma-4-26b-a4b-reasoning · How It Works · Data refreshed daily, snapshot 2026-08-05.