Alibaba Qwen · First MLX build 6 Aug 2025
Qwen3-4B-Instruct-2507
MLX
Text apache-2.0
A 4B parameter text generation model from Alibaba Qwen, with 13 MLX builds, from 4-bit to bf16.
- Parameters
- 4B From Hugging Face
- Quantisation
- 4-bit Qwen3-4B-Instruct-2507-4bit
- Memory needed
- 5 GB Estimate, unified memory
- Download
- 2.3 GB Weights, estimated
- Downloads
- 96K This build, last 30 days
- Task
- Text generation Text
- MLX builds
- 13 On mlx-community
- Devices
- Mac Apple silicon
Memory
Will it fit?Needs about 5 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/Qwen3-4B-Instruct-2507-4bit --prompt "Hello"MLX builds
13 on mlx-community| Qwen3-4B-Instruct-2507-DDWQ | 4-bit | 570 MB | 3 GB | 36 |
|---|---|---|---|---|
| Qwen3-4B-Instruct-2507-4bit | 4-bit | 2.3 GB | 5 GB | 96K |
| Qwen3-4B-Instruct-2507-4bit-DWQ-2510 | 4-bit DWQ | 2.3 GB | 5 GB | 1.1K |
| Qwen3-4B-Instruct-2507-nvfp4 | 4-bit | 2.3 GB | 5 GB | 177 |
| Qwen3-4B-Instruct-2507-4bit-g32 | 4-bit | 2.3 GB | 5 GB | 88 |
| Qwen3-4B-Instruct-2507-mxfp4 | MXFP4 | 2.3 GB | 5 GB | 79 |
| Qwen3-4B-Instruct-2507-5.0bpw-dynamic | 4-bit | 2.3 GB | 5 GB | 45 |
| Qwen3-4B-Instruct-2507-5bit | 5-bit | 2.8 GB | 6 GB | 156 |
| Qwen3-4B-Instruct-2507-5bit-g32 | 5-bit | 2.8 GB | 6 GB | 68 |
| Qwen3-4B-Instruct-2507-6bit | 6-bit | 3.3 GB | 6 GB | 215 |
| Qwen3-4B-Instruct-2507-8bit | 8-bit | 4.3 GB | 7 GB | 818 |
| Qwen3-4B-Instruct-2507-mxfp8 | MXFP8 | 4.3 GB | 7 GB | 161 |
| Qwen3-4B-Instruct-2507-bf16 | bf16 | 8.0 GB | 11 GB | 289 |
Sizes are estimated from Hugging Face’s parameter counts and the quantisation. Downloads are for the last 30 days.