---
title: Gemma 4 26B-A4B Apple Silicon Benchmarks
summary: Mixture-of-experts Gemma 4, 15.5 GB. As fast as E4B on stills. Slower on chunks, because it writes more tokens.
---

# Gemma 4 26B-A4B Apple Silicon Benchmarks

Mixture-of-experts Gemma 4, 15.5 GB. As fast as E4B on stills. Slower on chunks, because it writes more tokens.

- Repo: `mlx-community/gemma-4-26b-a4b-it-4bit`
- Maker: Google DeepMind
- Maker page: /resources/groundslate/benchmarks/makers/google-deepmind
- Line: Gemma 4
- Line page: /resources/groundslate/benchmarks/makers/google-deepmind/gemma-4
- License: [Apache License 2.0](https://ai.google.dev/gemma/docs/gemma_4_license)

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.

- Size: 15.5 GB
- Tools: list_models, set_vision_model, index_video, index_image, process_video, reindex_clips, name_video, summarize_video
- Suite: /resources/groundslate/benchmarks/vision

## On each Mac

| Machine | Status | Still | Video chunk |
| --- | --- | --- | --- |
| MacBook Pro, Apple M3 Max, 64 GB unified memory | Reference Mac | 3.13s | 11.83s |

## Speed

- Still: 3.13s, prefill 0.97s, 153 tokens
- Video chunk: 11.83s, 635 tokens