Model Details
Gemma 3n E4B LiteRT Preview
Quick answer: Gemma 3n E4B LiteRT Preview is Google's May 2025 on-device model scoring 61.6% ARC-C, 64.5% Global MMLU Lite, 78.6% HellaSwag. LiteRT-optimised for edge deployment.
At a Glance
Where Gemma 3n E4B LiteRT Preview leads
- 61.6% ARC-C — strong reasoning for E4B edge
- LiteRT: hardware-accelerated on-device inference
- Google ecosystem integration
Where it lags
- Preview: not production-ready
- E4B: compact scale limits capability
Best for: On-device Android/ChromeOS AI workloads; Google Pixel inference; LiteRT hardware acceleration.
What Gemma 3n E4B LiteRT Preview Is
Gemma 3n E4B LiteRT Preview is part of the Google May 2025 Gemma 3n family, optimised for LiteRT (formerly TensorFlow Lite) on-device inference. The E4B variant is smaller than the E4B, targeting the most constrained edge hardware.
Specifications
| Field | Value |
|---|---|
| Organization | |
| License | Gemma Terms of Use |
| HuggingFace | google/gemma-3n-e4b-litert-preview |
| Release date | May 20, 2025 |
| Modality | Text only |
Pricing
Open weights under Gemma ToU — self-host at no cost.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| ARC-C | 61.6% | Benchgen evaluation | 2025-05 |
| Global MMLU Lite | 64.5% | Benchgen evaluation | 2025-05 |
| HellaSwag | 78.6% | Benchgen evaluation | 2025-05 |
Gemma 3n E4B LiteRT vs Alternatives
| Model | ARC-C | Global MMLU Lite | License |
|---|---|---|---|
| Gemma 3n E4B LiteRT | 61.6% | 64.5% | Gemma ToU |
| Gemma 3n E4B | — | — | Gemma ToU |
| Gemma 3n E2B | — | — | Gemma ToU |
Related Models & Benchmarks
- Compare: Gemma 3n E4B · Gemma 3n E2B
- Benchmarks: ARC-C · Global MMLU Lite · HellaSwag
Frequently Asked Questions
What is Gemma 3n E4B LiteRT Preview?
Google's May 2025 on-device E4B preview scoring 61.6% ARC-C, 64.5% Global MMLU Lite. LiteRT edge inference.Specs from Google's Gemma 3n E4B LiteRT Preview release (May 2025) and Benchgen evaluations. Last updated 2026-07-24.