---
title: Gemma 4 Apple Silicon Benchmarks
summary: E2B and E4B are effective-parameter mixtures. The 12B is dense. 26B-A4B is a mixture-of-experts. GroundSlate runs the 4-bit MLX conversions from mlx-community.
---

# Gemma 4 Apple Silicon Benchmarks

E2B and E4B are effective-parameter mixtures. The 12B is dense. 26B-A4B is a mixture-of-experts. GroundSlate runs the 4-bit MLX conversions from mlx-community.

- Maker: Google DeepMind
- Maker page: /resources/groundslate/benchmarks/makers/google-deepmind
- Site: https://ai.google.dev/gemma

## Models

- Gemma 4 E2B (`mlx-community/gemma-4-e2b-it-4bit`): still 2.88s, video chunk 10.97s. /resources/groundslate/benchmarks/vision/gemma-4-e2b
- Gemma 4 E4B (`mlx-community/gemma-4-e4b-it-4bit`): still 3.49s, video chunk 7.86s. /resources/groundslate/benchmarks/vision/gemma-4-e4b
- Gemma 4 12B (`mlx-community/gemma-4-12B-it-4bit`): still 6.45s, video chunk 27.49s. /resources/groundslate/benchmarks/vision/gemma-4-12b
- Gemma 4 26B-A4B (`mlx-community/gemma-4-26b-a4b-it-4bit`): still 3.13s, video chunk 11.83s. /resources/groundslate/benchmarks/vision/gemma-4-26b-a4b