# Gemma 4 E4B ![Organization](https://img.shields.io/badge/Org-Google-4285F4) ![Effective Params](https://img.shields.io/badge/Effective-4B-1f6feb) ![License](https://img.shields.io/badge/License-Gemma%20ToU-d29922) ![Weights](https://img.shields.io/badge/Weights-Open-2da44e) ![On--Device](https://img.shields.io/badge/Target-On--Device-2da44e) ![Released](https://img.shields.io/badge/Released-Dec%202025-555) > **Quick answer:** Gemma 4 E4B is Google's December 2025 on-device multimodal model with 4B effective parameters, scoring 69.4% MMLU-Pro. It is the Gemma 4 generation equivalent of Gemma 3n E4B — updated architecture with a September 2025 knowledge cutoff and improved benchmark performance. ## At a Glance **Where Gemma 4 E4B leads** - 69.4% MMLU-Pro — strong academic knowledge at 4B effective param scale - On-device capable multimodal model - September 2025 knowledge cutoff — latest among on-device Gemma models - Gemma 4 generation improvements vs Gemma 3n E4B (50.6%) **Where it lags** - Gemma Terms of Use — commercial restrictions - 4B scale limits vs larger models - Limited benchmark coverage publicly available **Best for:** On-device multimodal deployment at latest Gemma generation; mobile apps requiring strong MMLU-Pro at 4B scale. ## What Gemma 4 E4B Is Gemma 4 E4B (released December 2025) is Google's on-device multimodal model in the Gemma 4 generation. It succeeds Gemma 3n E4B (May 2025, 50.6% MMLU-Pro) with significantly improved MMLU-Pro (69.4%) and a more recent September 2025 knowledge cutoff. The Gemma 4 on-device family continues Google's MatFormer-based architecture for efficient on-device inference, with text, image, and likely audio inputs maintained from the Gemma 3n generation. The +18.8pp MMLU-Pro improvement (50.6% → 69.4%) is a substantial generation-on-generation leap. ## Specifications | Field | Value | |---|---| | Organization | Google | | Effective parameters | 4B | | License | Gemma Terms of Use | | HuggingFace | google/gemma-4-E4B-it | | Release date | December 2025 | | Knowledge cutoff | September 2025 | | Modality | Multimodal | ## Pricing Open weights under Gemma Terms of Use — free to self-host. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [MMLU-Pro](/benchmarks/mmlu-pro) | 69.4% | Benchgen evaluation | 2025-12 | ## Gemma 4 E4B vs Alternatives | Model | MMLU-Pro | On-Device | License | |---|---|---|---| | **Gemma 4 E4B** | 69.4% | Yes | Gemma ToU | | [Gemma 3n E4B](/models/google/gemma-3n-e4b) | 50.6% | Yes | Gemma ToU | | [Gemma 4 E2B](/models/google/gemma-4-e2b) | 60% | Yes | Gemma ToU | Gemma 4 E4B vs Gemma 3n E4B: +18.8pp MMLU-Pro — significant improvement. vs Gemma 4 E2B: higher score (69.4% vs 60%) at higher inference cost. ## Related Models & Benchmarks - Compare: [Gemma 3n E4B](/models/google/gemma-3n-e4b) · [Gemma 4 E2B](/models/google/gemma-4-e2b) · [Gemma 4 26B A4B](/models/google/gemma-4-26b-a4b) - Benchmarks: [MMLU-Pro](/benchmarks/mmlu-pro) ## Frequently Asked Questions

What is Gemma 4 E4B?

Google's December 2025 on-device multimodal 4B effective parameter model scoring 69.4% MMLU-Pro. Significant improvement over Gemma 3n E4B (50.6%).

Should I use Gemma 4 E4B or Gemma 3n E4B?

Gemma 4 E4B for new deployments: +18.8pp MMLU-Pro, more recent knowledge cutoff (September 2025 vs February 2025).

--- *Specs from Google's Gemma 4 release (December 2025) and Benchgen evaluations. Last updated 2026-07-24.*

Gemma 4 E4B — Training jobs

Training jobs