Memory at each precision
| Version | Weights | With context | Smallest Mac |
|---|---|---|---|
| 8-bit | 30.6 GB | 33 GB | 48 GB |
| 6-bit | 23.4 GB | 25.3 GB | 36 GB |
| 4-bit | 16.2 GB | 17.7 GB | 24 GB |
| 3-bit lower quality | 12.6 GB | 14.1 GB | 24 GB |
| 2-bit lower quality | 9 GB | 10.5 GB | 16 GB |
On every Apple chip
The smallest memory size of each chip that runs Raon-1 mini Preview, 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 | 32–57 |
| M5 Ultra | 96 GB | 8-bit Runs well | 117–209 |
| M5 Max | 36 GB | 6-bit Runs well | 59–105 |
| M5 Pro | 24 GB | 4-bit Runs well | 58–102 |
| M5 | 24 GB | 4-bit Runs well | 29–51 |
| M4 Max | 36 GB | 6-bit Runs well | 53–94 |
| M4 Pro | 24 GB | 4-bit Runs well | 51–91 |
| M4 | 24 GB | 4-bit Runs well | 23–40 |
| M3 Ultra | 96 GB | 8-bit Runs well | 80–142 |
| M3 Max | 36 GB | 6-bit Runs well | 39–69 |
| M3 Pro | 36 GB | 6-bit Runs well | 19–34 |
| M3 | 24 GB | 4-bit Runs well | 19–33 |
| M2 Ultra | 64 GB | 8-bit Runs well | 78–139 |
| M2 Max | 32 GB | 4-bit Runs well | 75–133 |
| M2 Pro | 32 GB | 4-bit Runs well | 38–67 |
| M2 | 24 GB | 4-bit Runs well | 19–33 |
| M1 Ultra | 64 GB | 8-bit Runs well | 78–139 |
| M1 Max | 32 GB | 4-bit Runs well | 75–133 |
| M1 Pro | 32 GB | 4-bit Runs well | 38–67 |
| M1 | 16 GB | 2-bit Runs well | 24–42 |
How to run it
pip install mlx-lmmlx_lm.chat --model KRAFTON/Raon-1-mini-preview-mlx --trust-remote-code--trust-remote-code lets mlx-lm run the Python code that ships with this build, which it needs to load.Build: KRAFTON/Raon-1-mini-preview-mlx on Hugging Face. Other precisions sit next to it.Step by step: download and run LLMs on a Mac · MLX vs Ollama vs llama.cpp
Prefer an app to the Terminal? Compare the Mac apps that run local models in the AI App Store: price, licence and where your data goes.
About Raon-1 mini Preview
| Made by | KRAFTON |
|---|---|
| Parameters | 28.8B total · 4.3B active (mixture of experts) |
| Context | 256K tokens |
| Licence | Apache-2.0 |
| Released | 2026-10 |
| Input | Text |
| Weights | KRAFTON/Raon-1-mini-preview |
Questions
- How much memory does Raon-1 mini Preview need?
- About 16.2 GB of weights at 4-bit, plus room for the context: 17.7 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 Raon-1 mini Preview on a Mac?
- Writing speed depends on memory bandwidth. The fastest Mac for it here is an M5 Ultra with 96 GB at about 117–209 tokens a second (8-bit). The smallest Mac that runs it, with 24 GB, manages about 58–102 on an M5 Pro.
- Why does Raon-1 mini Preview run faster than its size suggests?
- It is a mixture-of-experts model: it holds 28.8B parameters in memory but reads only 4.3B 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 Raon-1 mini Preview well and links here.
[](https://mlx.app/run/raon-1-mini-preview)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.
Open data: mlx.app/data (JSON, CSV)