Model Details
Gemma 4 E4B
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 | |
| 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 | 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 | 50.6% | Yes | Gemma ToU |
| 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 · Gemma 4 E2B · Gemma 4 26B A4B
- 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.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.