Gemma 4 E2B
Gemma 4 E2B is Google DeepMind's smallest multimodal model with 2.3 billion effective parameters (5.1B with embeddings) and a 128K context window. Supports image, text, and audio inputs. Designed for on-device and edge deployment with Per-Layer Embeddings for efficient inference.
Benchmark results
| Benchmark | Score | Tags | Source |
|---|---|---|---|
| AIME 2026 | 37.5% | self-reported llm-stats | link → |
| BIG-Bench Extra Hard | 21.9% | self-reported llm-stats | link → |
| GPQA | 43.4% | self-reported llm-stats | link → |
| LiveCodeBench v6 | 44.0% | self-reported llm-stats | link → |
| MathVision | 52.4% | self-reported llm-stats | link → |
| MedXpertQA | 23.5% | self-reported llm-stats | link → |
| MMLU-Pro | 60.0% | self-reported llm-stats | link → |
| MMMLU | 67.4% | self-reported llm-stats | link → |
| MMMU-Pro | 44.2% | self-reported llm-stats | link → |
| MRCR v2 | 19.1% | self-reported llm-stats | link → |
| t2-bench | 29.4% | self-reported llm-stats | link → |