Memory at each precision
| Version | Weights | With context | Smallest Mac |
|---|---|---|---|
| 8-bit | 26.8 GB | 28.9 GB | 48 GB |
| 6-bit | 20.5 GB | 22.1 GB | 32 GB |
| 4-bit | 14.2 GB | 15.7 GB | 24 GB |
| 3-bit lower quality | 11 GB | 12.5 GB | 18 GB |
| 2-bit lower quality | 7.9 GB | 9.4 GB | 16 GB |
On every Apple chip
The smallest memory size of each chip that runs Gemma 4 26B A4B, the best version that fits it, and the writing speed you can expect.
| Chip | Needs | Version | Tokens/s |
|---|---|---|---|
| M6 | 24 GB | 4-bit Runs well | 36–65 |
| M5 Ultra | 96 GB | 8-bit Runs well | 135–240 |
| M5 Max | 36 GB | 6-bit Runs well | 67–119 |
| M5 Pro | 24 GB | 4-bit Runs well | 66–117 |
| M5 | 24 GB | 4-bit Runs well | 33–58 |
| M4 Max | 36 GB | 6-bit Runs well | 60–106 |
| M4 Pro | 24 GB | 4-bit Runs well | 59–104 |
| M4 | 24 GB | 4-bit Runs well | 26–46 |
| M3 Ultra | 96 GB | 8-bit Runs well | 92–164 |
| M3 Max | 36 GB | 6-bit Runs well | 44–77 |
| M3 Pro | 36 GB | 6-bit Runs well | 22–39 |
| M3 | 24 GB | 4-bit Runs well | 21–38 |
| M2 Ultra | 64 GB | 8-bit Runs well | 90–160 |
| M2 Max | 32 GB | 6-bit Runs well | 58–103 |
| M2 Pro | 32 GB | 6-bit Runs well | 29–52 |
| M2 | 24 GB | 4-bit Runs well | 21–38 |
| M1 Ultra | 64 GB | 8-bit Runs well | 90–160 |
| M1 Max | 32 GB | 6-bit Runs well | 58–103 |
| M1 Pro | 32 GB | 6-bit Runs well | 29–52 |
| M1 | 16 GB | 2-bit Runs well | 26–45 |
How to run it
- The easy way: download MLX Harness, free and open source. Search for Gemma 4 26B A4B, download the version that fits your Mac, and chat. It also serves the model to your other apps over an OpenAI-compatible API.Download MLX HarnessFree · Apple Silicon · macOS 14.5 or later
- In Terminal with Apple’s MLX:
pip install mlx-lmmlx_lm.chat --model mlx-community/gemma-4-26b-a4b-it-4bitBuild: mlx-community/gemma-4-26b-a4b-it-4bit on Hugging Face. Other precisions sit next to it.
Step by step: download and run LLMs on a Mac · how to use MLX Harness · MLX vs Ollama vs llama.cpp
About Gemma 4 26B A4B
| Made by | Google DeepMind |
|---|---|
| Parameters | 25.2B total · 3.8B active (mixture of experts) |
| Context | 256K tokens |
| Licence | Apache-2.0 |
| Released | 2026-03 |
| Input | Text and images |
| Weights | google/gemma-4-26B-A4B-it |
Questions
- How much memory does Gemma 4 26B A4B need?
- About 14.2 GB of weights at 4-bit, plus room for the context: 15.7 GB in all. On a Mac that means 24 GB of unified memory or more, since macOS lets the GPU use about 75% of it by default.
- How fast is Gemma 4 26B A4B on a Mac?
- Writing speed depends on memory bandwidth. The fastest Mac for it here is an M5 Ultra with 96 GB at about 135–240 tokens a second (8-bit). The smallest Mac that runs it, with 24 GB, manages about 66–117 on an M5 Pro.
- Why does Gemma 4 26B A4B run faster than its size suggests?
- It is a mixture-of-experts model: it holds 25.2B parameters in memory but reads only 3.8B for each token, so it writes at the speed of a much smaller model.
Similar models
For your model card or README
Shows the smallest Mac that runs Gemma 4 26B A4B well and links here.
[](https://mlx.app/run/gemma-4-26b-a4b)How it’s worked out
Memory: the model’s parameters at 8, 6, 4, 3 or 2 bits per weight plus MLX’s group scales (half a bit per weight), or its native format, with room for the context on top. By default macOS lets the GPU use about 75% of unified memory. Speed: memory bandwidth divided by the bytes read per token is the ceiling for one conversation; in llama.cpp’s public Apple Silicon benchmarks real runs reach about 45% to 80% of it, which is the range shown. These are estimates, not measurements: measure your own Mac on MLX Bench. Model figures from each official model card, chip figures from Apple’s Tech Specs. Checked 2026-10-19. Same data as mlx.app/compare.