BAAI AREX 2 needs about 19.9 GB of memory at Q4_K_M (a 17.2 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 26 quants, from 0.0 GB to 53.8 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 8.6 GB.
BAAI AREX 2 – 27B multimodal LLM with long‑horizon self‑improvement
BAAI AREX 2 is a 27‑billion‑parameter model built on the qwen35 architecture. It supports text and image inputs and is trained for long‑horizon reasoning and self‑improvement tasks. The model offers a 262,144‑token context window and is released under the Apache‑2.0 license.
Good at
- Long‑horizon self‑improvement
- Multimodal reasoning with image input
- Tool‑aware task execution
Limits
- Large VRAM required for higher‑quality quants
- No speculative decoding support
- Requires llama.cpp b11259 or newer
Best for
- Iterative coding and debugging workflows
- Research agents needing extended context
- Multimodal content generation
Who it’s for. Developers and researchers who need a large, multimodal LLM capable of iterative refinement and long‑horizon reasoning. The model is suitable for building agentic systems that can process and analyze large documents and images.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 0.0 GB | 2.7 GB | Fits | Get file → | |
| IQ2_XXS | 8.6 GB | 11.3 GB | Too big | Get file → | |
| IQ2_XS | 8.8 GB | 11.5 GB | Too big | Get file → | |
| IQ2_S | 9.4 GB | 12.1 GB | Too big | Get file → | |
| IQ2_M | 10.3 GB | 12.9 GB | Too big | Get file → | |
| Q2_K | 10.6 GB | 13.2 GB | Too big | Get file → | |
| IQ3_XXS | 12.1 GB | 14.7 GB | Too big | Get file → | |
| Q3_K_S | 12.5 GB | 15.1 GB | Too big | Get file → | |
| IQ3_XS | 12.6 GB | 15.2 GB | Too big | Get file → | |
| Q3_K_M | 13.2 GB | 15.8 GB | Too big | Get file → | |
| Q3_K_L | 13.9 GB | 16.5 GB | Too big | Get file → | |
| IQ3_M | 14.6 GB | 17.3 GB | Too big | Get file → | |
| IQ4_XS | 15.2 GB | 17.9 GB | Too big | Get file → | |
| Q4_0 | 16.1 GB | 18.8 GB | Too big | Get file → | |
| Q4_K_S | 16.1 GB | 18.8 GB | Too big | Get file → | |
| IQ4_NL | 17.2 GB | 19.9 GB | Too big | Get file → | |
| Q4_K_M | 17.2 GB | 19.9 GB | Too big | Get file → | |
| Q4_1 | 17.6 GB | 20.2 GB | Too big | Get file → | |
| Q4_K_L | 18.6 GB | 21.2 GB | Too big | Get file → | |
| Q5_K_S | 19.3 GB | 22.0 GB | Too big | Get file → | |
| Q5_K_M | 20.7 GB | 23.3 GB | Too big | Get file → | |
| Q6_K_S | 22.6 GB | 25.3 GB | Too big | Get file → | |
| Q6_K | 23.6 GB | 26.3 GB | Too big | Get file → | |
| Q6_K_L | 24.7 GB | 27.4 GB | Too big | Get file → | |
| Q8_0 | 28.7 GB | 31.3 GB | Too big | Get file → | |
| BF16 | 53.8 GB (2 parts) | 56.5 GB | Too big | Get file → |
For 8–12 GB VRAM, use IQ2_XXS or IQ2_XS (≈11 GB memory at 8k context). For 16–20 GB VRAM, Q4_K_M or Q4_K_S (≈19–20 GB). For 24–30 GB VRAM, Q6_K or Q6_K_L (≈26–27 GB). The estimates are approximate and depend on the exact context length used.
What GPU do you need for BAAI AREX 2?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 2.7 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 2.7 GB in use).
- 8 GB VRAM: the best quant that fits is the main file (0.0 GB file, about 2.7 GB in use).
- 12 GB VRAM: the best quant that fits is IQ2_XS (8.8 GB file, about 11.5 GB in use).
- 16 GB VRAM: the best quant that fits is Q3_K_M (13.2 GB file, about 15.8 GB in use).
- 24 GB VRAM: the best quant that fits is Q5_K_M (20.7 GB file, about 23.3 GB in use).
- 32 GB VRAM: the best quant that fits is Q8_0 (28.7 GB file, about 31.3 GB in use).
- 48 GB VRAM: the best quant that fits is Q8_0 (28.7 GB file, about 31.3 GB in use).
Which BAAI AREX 2 quant should you choose?
Q4_K_M is the usual starting point for BAAI AREX 2: it balances file size and quality well. Here it is a 17.2 GB file.
The Q8_0 file is 28.7 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is the first one in the table at 0.0 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about BAAI AREX 2
How much VRAM does BAAI AREX 2 need?
About 19.9 GB at Q4_K_M with an 8K context window. That is the 17.2 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run BAAI AREX 2 on an 8 GB GPU?
Yes, with the smaller files: the 0.0 GB file needs about 2.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of BAAI AREX 2 GGUF?
It depends on the quant: file is 0.0 GB, IQ2_XXS is 8.6 GB, IQ2_XS is 8.8 GB, IQ2_S is 9.4 GB, IQ2_M is 10.3 GB, Q2_K is 10.6 GB and 20 more.
What context length does BAAI AREX 2 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 2.1 GB at 8K and 8.6 GB at 32K.
Where can I download BAAI AREX 2 GGUF?
The files are on Hugging Face in the bartowski/BAAI_AREX-2-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does BAAI AREX 2 use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run BAAI AREX 2 on 8 GB VRAM?
Only the smallest quants (IQ2_XXS or IQ2_XS) fit within 8 GB VRAM, but they still require around 11 GB of memory at 8k context, so 8 GB is insufficient.
Does BAAI AREX 2 support vision tasks?
Yes, the model supports image input via the mmproj projector files included in the repo.
What license applies to BAAI AREX 2?
The model is released under the Apache‑2.0 license, allowing commercial and non‑commercial use.