4-bit Open on Hugging Face
Parameters
27B
Weights (est.)
≈ 16.1 GB
Mac memory (est.)
24 GB+
Size is estimated from the parameter count and quantisation. Memory allows a fifth extra for context plus 2 GB for the system; long contexts need more.
Run it
With mlx-vlm on an Apple silicon Mac:
pip install mlx-vlmpython -m mlx_vlm.generate --model mlx-community/AREX-2-4bit --image photo.jpg --prompt "Describe this image"
Details
- Base model
- BAAI/AREX-2 (quantized)
- Vendor
- BAAI
- Task
- Vision + text
- Quantisation
- 4-bit
- Licence
- apache-2.0
- Downloads
- 218 in the last 30 days · 355 all time · 0 likes
- Published
- 2026-10-06 · updated 2026-10-07 (1mo ago)
- Languages
- en
- Tags
- qwen3_5, apple-silicon, quantized, arex-2, qwen3.8, vision, agent, reasoning, tool-use, deep-research, long-context
Other MLX builds of BAAI/AREX-2
| Build | Quantisation | Weights | Mac memory | Downloads |
|---|---|---|---|---|
| AREX-2-4bit (this one) | 4-bit | ≈ 16.1 GB | 24 GB+ | 218 |
| AREX-2-5bit | 5-bit | ≈ 19.4 GB | 24 GB+ | 299 |
| AREX-2-6bit | 6-bit | ≈ 22.8 GB | 32 GB+ | 379 |
| AREX-2-8bit | 8-bit | ≈ 29.5 GB | 48 GB+ | 181 |