Gemma 3n E2B Instructed LiteRT (Preview)

Gemma 3n is a generative AI model optimized for use in everyday devices, such as phones, laptops, and tablets. It features innovations like Per-Layer Embedding (PLE) parameter caching and a MatFormer model architecture for reduced compute and memory. These models handle audio, text, and visual data, though this E4B preview currently supports text and vision input. Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models, and is licensed for responsible commercial use.

Benchmark results

Benchmark Score Tags Source
AIME 2025 6.7% self-reported llm-stats link →
ARC-C 51.7% self-reported llm-stats link →
ARC-E 75.8% self-reported llm-stats link →
BIG-Bench Hard 44.3% self-reported llm-stats link →
BoolQ 76.4% self-reported llm-stats link →
Codegolf v2.2 11.0% self-reported llm-stats link →
DROP 53.9% self-reported llm-stats link →
ECLeKTic 2.5% self-reported llm-stats link →
Global-MMLU 55.1% self-reported llm-stats link →
Global-MMLU-Lite 59.0% self-reported llm-stats link →
GPQA 24.8% self-reported llm-stats link →
HellaSwag 72.2% self-reported llm-stats link →
HiddenMath 27.7% self-reported llm-stats link →
HumanEval 66.5% self-reported llm-stats link →
Include 38.6% self-reported llm-stats link →
LiveCodeBench 13.2% self-reported llm-stats link →
LiveCodeBench v5 18.6% self-reported llm-stats link →
MBPP 56.6% self-reported llm-stats link →
MGSM 53.1% self-reported llm-stats link →
MMLU 60.1% self-reported llm-stats link →
MMLU-Pro 40.5% self-reported llm-stats link →
MMLU-ProX 8.1% self-reported llm-stats link →
Natural Questions 15.5% self-reported llm-stats link →
PIQA 78.9% self-reported llm-stats link →
Social IQa 48.8% self-reported llm-stats link →
TriviaQA 60.8% self-reported llm-stats link →
Winogrande 66.8% self-reported llm-stats link →
WMT24++ 42.7% self-reported llm-stats link →