Moonshot AI · First MLX build 31 Oct 2025
Kimi-Linear-48B-A3B-Instruct
MLX
Text mit
A 49B parameter text generation model from Moonshot AI, with 6 MLX builds, from 3-bit to bf16.
- Parameters
- 49B From Hugging Face
- Quantisation
- 4-bit Kimi-Linear-48B-A3B-Instruct-4bit
- Memory needed
- 33 GB Estimate, unified memory
- Download
- 27.6 GB Weights, estimated
- Downloads
- 652 This build, last 30 days
- Task
- Text generation Text
- MLX builds
- 6 On mlx-community
- Devices
- Mac Apple silicon
Memory
Will it fit?Needs about 33 GB at 4-bit
8 GB
16 GB
24 GB
32 GB
64 GB
128 GB
Fits with headroomTight, close other appsWon’t fit
Estimated from the size of the weights, plus a fifth for context and working memory and 2 GB for macOS and your apps. Long contexts need more.
Run it
Terminal · mlx-lm$ pip install -U mlx-lm
$ mlx_lm.generate --model mlx-community/Kimi-Linear-48B-A3B-Instruct-4bit --prompt "Hello"MLX builds
6 on mlx-community| Kimi-Linear-48B-A3B-Instruct-3bit | 3-bit | 21.5 GB | 27 GB | 106 |
|---|---|---|---|---|
| Kimi-Linear-48B-A3B-Instruct-4bit | 4-bit | 27.6 GB | 33 GB | 652 |
| Kimi-Linear-48B-A3B-Instruct-5bit | 5-bit | 33.8 GB | 40 GB | 102 |
| Kimi-Linear-48B-A3B-Instruct-6bit | 6-bit | 39.9 GB | 47 GB | 138 |
| Kimi-Linear-48B-A3B-Instruct-8bit | 8-bit | 52.2 GB | 61 GB | 121 |
| Kimi-Linear-48B-A3B-Instruct-mlx-bf16 | bf16 | 98.2 GB | 112 GB | 70 |
Sizes are estimated from Hugging Face’s parameter counts and the quantisation. Downloads are for the last 30 days.