Gemma 4 31B (Reasoning): benchmark results
Gemma 4 31B evaluated with reasoning enabled. Provider: Google. Released 2026-04-02. Access: Open.
Aggregate Rating 1627 ± 12, rank #616 of 2168 rated models, from 116 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Meta Muse Glimmer 30B Card - CIMemories - Violation Rate | 12.1 | Violation Rate (%, lower is better) | 100 |
| Meta Muse Glimmer 30B Card - Siren AgentDojo - Attack Success Rate | 25.6 | Attack Success Rate (%, lower is better) | 100 |
| AA IFBench | 75.58 | Accuracy (%) | 93.4 |
| AgentVidBench (Single-Turn) | 37 | Accuracy (%; 100 human-authored multi-hop multiple-choice qu | 84.2 |
| AA Terminal-Bench Hard | 36.36 | Accuracy (%) | 83.3 |
| AA GPQA Diamond | 85.66 | Accuracy (%) | 79.4 |
| PACT (Lechmazur) | 1557 | PACT Bilateral Rating | 74 |
| AA Omniscience - Software Engineering (SWE) - PHP | 30 | Accuracy (%) | 72.5 |
| AA Omniscience - Software Engineering (SWE) - Dart | 24 | Accuracy (%) | 72.2 |
| AA Humanity's Last Exam | 23.63 | Accuracy (%) | 70.3 |
| AA Omniscience - Software Engineering (SWE) - JavaScript | 33.64 | Accuracy (%) | 69.1 |
| AA Omniscience - Software Engineering (SWE) - Go | 22 | Accuracy (%) | 68.4 |
Interactive version: theaggregate.ai/model?slug=gemma-4-31b-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.