Holo4 35B A3B needs about 21.0 GB of memory at MXFP4_MOE (a 19.8 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 2 quants, from 19.8 GB (MXFP4_MOE) to 69.4 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.7 GB.
Holo4-35B-A3B: 35B MoE vision‑language model for computer‑use agents
Holo4-35B-A3B is a 35B MoE vision‑language model built on Qwen3.6-35B-A3B. It supports screenshot‑based UI interaction, tool calls, and code execution. The model has a 262,144 token context window and is released under Apache‑2.0.
Good at
- Vision‑language reasoning
- Computer‑use tool execution
- Large context handling
Limits
- No native vision support in llama.cpp
- Large VRAM requirement for BF16
- Limited to 262,144 token context
Best for
- Automated UI testing
- Code generation from screenshots
- Desktop automation scripts
Who it’s for. Developers building AI agents that interact with graphical interfaces. Researchers needing a large‑context vision‑language model for computer‑use tasks.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| MXFP4_MOE | 19.8 GB | 21.0 GB | Too big | Get file → | |
| BF16 | 69.4 GB | 70.5 GB | Too big | Get file → |
For 8 GB VRAM, use MXFP4_MOE (19.79 GB file, 21 GB memory at 8 k context). For 16 GB VRAM, BF16 (69.38 GB file, 70.5 GB memory at 8 k context) is available but requires more memory. Choose the quant that fits your GPU budget.
What GPU do you need for Holo4 35B A3B?
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: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: no quant fits at an 8K context.
- 12 GB VRAM: no quant fits at an 8K context.
- 16 GB VRAM: no quant fits at an 8K context.
- 24 GB VRAM: the best quant that fits is MXFP4_MOE (19.8 GB file, about 21.0 GB in use).
- 32 GB VRAM: the best quant that fits is MXFP4_MOE (19.8 GB file, about 21.0 GB in use).
- 48 GB VRAM: the best quant that fits is MXFP4_MOE (19.8 GB file, about 21.0 GB in use).
Which Holo4 35B A3B quant should you choose?
MXFP4_MOE (19.8 GB) is the middle option in this repository. Lower bit counts mean smaller files and more quality loss.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Holo4 35B A3B
How much VRAM does Holo4 35B A3B need?
About 21.0 GB at MXFP4_MOE with an 8K context window. That is the 19.8 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 Holo4 35B A3B on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 21.0 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Holo4 35B A3B GGUF?
It depends on the quant: MXFP4_MOE is 19.8 GB, BF16 is 69.4 GB.
What context length does Holo4 35B A3B support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.7 GB at 8K and 2.7 GB at 32K.
Where can I download Holo4 35B A3B GGUF?
The files are on Hugging Face in the prithivMLmods/Holo4-35B-A3B-MXFP4-MoE-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 Holo4 35B A3B 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 Holo4-35B-A3B on 8 GB VRAM?
Only the MXFP4_MOE quant is feasible; it needs 21 GB memory at 8 k context, so 8 GB VRAM is insufficient.
Does this model support vision in llama.cpp?
No, the model has vision capabilities but llama.cpp does not expose them.
What license applies to this model?
The model is released under Apache‑2.0.