Gemma 4 E4B (IT) (Non-reasoning): benchmark results
Provider: Google. Released 2026-04-02. Access: Open.
Unified ELO 1478 ± 1, rank #1674 of 3078 rated models, from 28 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| FrameBench - Japanese | 84.6 | Accuracy (%; mean of five prompt templates) | 52 |
| MERA v2 - GorillaHard | 20.9 | Score (%) | 44.7 |
| MERA v2 - Characters | 42.7 | Score (%) | 36.8 |
| MERA v2 - NewReasoning | 37.7 | Score (%) | 36.8 |
| MERA v2 - Reasoning | 36.7 | Score (%) | 36.8 |
| FrameBench - Frame Identification - Japanese | 68.5 | Accuracy (%; Japanese FrameNet candidate frames) | 36 |
| MERA v2 - Agentic | 6.7 | Score (%) | 35.5 |
| MERA v2 | 0.17 | Total score | 34.2 |
| MERA v2 - Enantiosemy | 28.3 | Score (%) | 34.2 |
| MERA v2 - IFHardBench | 6.6 | Score (%) | 34.2 |
| MERA v2 - SOB-Hard | 2.1 | Score (%) | 34.2 |
| FrameBench - English | 82.5 | Accuracy (%; mean of five prompt templates) | 31.6 |
Interactive version: theaggregate.ai/model?slug=gemma-4-e4b-it-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-19.