Gemma 4 E2B: benchmark results
Google DeepMind's Gemma 4 E2B, a multimodal (text, image, audio) model in the five-size Gemma 4 family, with a 256K-token context, support for 140+ languages and configurable reasoning modes. Provider: Google. Released 2026-04-02. Access: Open.
Aggregate Rating 1436 ± 11, rank #1070 of 1656 rated models, from 469 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Jais 3 Model Cards - SimpleQA Verified (Arabic) - Non-hallucination Rate | 84.8 | Non-hallucination rate (%) | 100 |
| POLAR-Bench - Utility | 87.56 | Utility score (%): share of the task-relevant attributes the | 95.2 |
| QIMMA - ArabLegalQA | 72.39 | BERTScore F1 (%) | 93.2 |
| AA-Omniscience Hallucination Rate | 32.36 | Hallucination Rate (%) | 90.8 |
| aiXamine - Safety Alignment | 97.7 | Service score (0-100; normalized pass rate) | 82.9 |
| Bangladesh Legal QA (No Law) | 70 | Accuracy (%; no legal context supplied, parametric knowledge | 80 |
| Bangladesh Legal QA (Supplied Law) | 84.7 | Accuracy (%; the governing statutory section is supplied wit | 80 |
| PRISM-VLM - Hallucination Resistance | 71 | Per-axis mean score (%; share of trap questions (absent enti | 79.3 |
| aiXamine - Hallucination | 71.2 | Service score (0-100; normalized pass rate) | 65.9 |
| MathVision | 52.4 | Overall Accuracy (%) | 65 |
| AA-Omniscience Index - Software Engineering (SWE) - Dart | -16 | Omniscience Index | 61.6 |
| AA-Omniscience Index - Health | -22.17 | Omniscience Index | 61 |
Interactive version: theaggregate.ai/model?slug=gemma-4-e2b · How It Works · Data refreshed daily, snapshot 2026-10-20.