Gemma 4 31B (Non-reasoning): benchmark results
Provider: Google. Released 2026-04-02. Access: Open.
Aggregate Rating 1625 ± 17, rank #627 of 2170 rated models, from 59 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AMIGO (Threshold 0.3) | 25.2 | Verified Correct accuracy in percent on AMIGO's dress galler | 100 |
| AMIGO (Threshold 0.4) | 27.9 | Verified Correct accuracy in percent on AMIGO's dress galler | 100 |
| AMIGO (Threshold 0.5) | 26.6 | Verified Correct accuracy in percent on AMIGO's dress galler | 100 |
| AMIGO (Threshold 0.6) | 18.6 | Verified Correct accuracy in percent on AMIGO's dress galler | 100 |
| TRACE Bench - Character Consistency | 95.46 | Checklist completion rate (%) | 100 |
| TRACE Bench - Overall | 96.02 | Weighted overall score (%) | 100 |
| TRACE Bench - Diversity | 93.49 | Normalized repetition-penalty score (%) | 96 |
| TRACE Bench - Length | 98.49 | Reply-length compliance (%) | 96 |
| VAMPS | 89.5 | Accuracy (%) on the English versions of the 218 Konkour-seed | 94.1 |
| VAMPS (Analytical) - Persian | 97.2 | Accuracy (%) on the original Persian versions of the 218 Kon | 94.1 |
| TRACE Bench - Short-Term Memory | 90.5 | Memory-probe completion rate (%) | 88 |
| VAMPS (Analytical) | 97.2 | Accuracy (%) on the English versions of the 218 Konkour-seed | 85.3 |
Interactive version: theaggregate.ai/model?slug=gemma-4-31b-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-05.