Gemma 4 E2B (IT): benchmark results
Google DeepMind's Gemma 4 E2B (IT), a small instruction-tuned multimodal model handling text, image and audio input, with configurable reasoning ('thinking') modes and up to 256K context. Provider: Google. Released 2026-04-02. Access: Open.
Aggregate Rating 1499 ± 8, rank #781 of 1656 rated models, from 819 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Omni-Persona - Audio-to-Audio | 31.2 | Calibrated accuracy (%): item-weighted accuracy over answera | 100 |
| Nejumi 4 - jaster (0-shot) - JSICK | 87 | Exact match (%) | 99.7 |
| EuroEval Italian Summarization - Ilpost SUM | 38.97 | chrF3++ (%) | 98.7 |
| Kurakurai French LLM - MBPP | 68.68 | pass@1 (%) | 97.6 |
| Kurakurai French LLM - MMLU-Pro | 56.4 | Accuracy (%) | 97.6 |
| Kurakurai French LLM - BFCL v2 | 72.98 | Accuracy (%) | 96.6 |
| Nejumi 4 - BFCL - Irrelevance Detection | 95 | Accuracy (%) | 96.5 |
| Kurakurai French LLM - AIME 2024 | 31.67 | Accuracy (%) | 95.1 |
| Kurakurai French LLM - AIME 2025 | 23 | Accuracy (%) | 95.1 |
| Kurakurai French LLM - Global MMLU | 67.8 | Accuracy (%) | 95.1 |
| Kurakurai French LLM - HumanEval+ | 79.06 | pass@1 (%) | 95.1 |
| Kurakurai French LLM - MGSM | 89.12 | Accuracy (%) | 95.1 |
Interactive version: theaggregate.ai/model?slug=gemma-4-e2b-it · How It Works · Data refreshed daily, snapshot 2026-10-20.