Gemma 3n E4B-it is optimized for efficient execution on mobile and low-resource devices, such as phones, laptops, and tablets. It supports multimodal inputs—including text, visual data, and audio—enabling diverse tasks such as text generation, speech recognition, translation, and image analysis. Leveraging innovations like Per-Layer Embedding (PLE) caching and the MatFormer architecture, Gemma 3n dynamically manages memory usage and computational load by selectively activating model parameters, significantly reducing runtime resource requirements.
This model supports a wide linguistic range (trained in over 140 languages) and features a flexible 32K token context window. Gemma 3n can selectively load parameters, optimizing memory and computational efficiency based on the task or device capabilities, making it well-suited for privacy-focused, offline-capable applications and on-device AI solutions. [Read more in the blog post](https://developers.googleblog.com/en/introducing-gemma-3n/)
Evaluations
26
across 18 benchmarks
Latency
441ms
Context length
33k
tokens
Cost
$0.02 · $0.04
input · output per 1M tokens
Key takeaways
Gemma 3n E4B-it is optimized for efficient execution on mobile and low-resource devices, supporting multimodal inputs (text, visual, audio) for diverse tasks like text generation, speech recognition, translation, and image analysis.
This Transformer-based model optimizes memory and computational efficiency using Per-Layer Embedding (PLE) caching and the MatFormer architecture, dynamically activating parameters.
The model produces human-like, non-toxic, and readable text, but its severe weakness in basic fact retrieval leads to frequent incorrect yet confidently delivered answers across various factual domains.
The model demonstrates strong performance in easy scientific reasoning, achieving 93.5% accuracy on 'AI2 Reasoning Challenge - Easy' with consistent, readable, and non-toxic responses.
A significant limitation is the model's struggle with factual recall, achieving only 4.02% accuracy on 'SimpleQA', often hallucinating confidently incorrect information and misinterpreting question nuances, especially in art trivia.
