Carnice Qwen3.6 MoE 35B A3B APEX MTP needs about 24.7 GB of memory at its main file (a 23.5 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 9 quants, from 11.7 GB to 71.1 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 2.8 GB.
Carnice Qwen3.6 MoE 35B A3B APEX MTP – a large‑scale MoE model with self‑speculative decoding
Carnice Qwen3.6 MoE 35B A3B APEX MTP is a 36‑billion‑parameter model with 3 billion active experts. It supports a 262 k token context and includes a bundled MTP head for self‑speculative decoding. The model is released under Apache‑2.0 and is optimized for agentic workflows.
Good at
- Agentic reasoning
- Long‑context handling
- Self‑speculative decoding
Limits
- Large VRAM requirement (≈12–72 GB for 8 k context)
- No vision support
- Requires recent llama.cpp (≥commit 255582687)
Best for
- Building Hermes‑compatible agents
- Long‑form coding assistance
- Complex multi‑step tool‑chain tasks
Who it’s for. Developers and researchers building autonomous agents who need a large MoE model with native support for self‑speculative decoding. The model is suitable for environments with high VRAM and recent llama.cpp builds.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 11.7 GB | 12.9 GB | Too big | Get file → | |
| — | 14.3 GB | 15.5 GB | Too big | Get file → | |
| — | 17.3 GB | 18.5 GB | Too big | Get file → | |
| — | 17.3 GB | 18.5 GB | Too big | Get file → | |
| — | 23.5 GB | 24.7 GB | Too big | Get file → | |
| — | 23.5 GB | 24.7 GB | Too big | Get file → | |
| — | 26.1 GB | 27.2 GB | Too big | Get file → | |
| — | 26.1 GB | 27.2 GB | Too big | Get file → | |
| F16 | 71.1 GB | 72.3 GB | Too big | Get file → |
For 8 GB VRAM use the 11.69 GB GGUF (≈12.9 GB memory). For 16 GB VRAM use the 14.27 GB file (≈15.5 GB memory). For 24 GB VRAM use the 23.48 GB file (≈24.7 GB memory). The 71.07 GB F16 quant is only for very large GPUs.
What GPU do you need for Carnice Qwen3.6 MoE 35B A3B APEX MTP?
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: the best quant that fits is the main file (14.3 GB file, about 15.5 GB in use).
- 24 GB VRAM: the best quant that fits is the main file (17.3 GB file, about 18.5 GB in use).
- 32 GB VRAM: the best quant that fits is the main file (26.1 GB file, about 27.2 GB in use).
- 48 GB VRAM: the best quant that fits is the main file (26.1 GB file, about 27.2 GB in use).
Which Carnice Qwen3.6 MoE 35B A3B APEX MTP quant should you choose?
If memory is tight, the smallest file is the first one in the table at 11.7 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
Frequently asked questions about Carnice Qwen3.6 MoE 35B A3B APEX MTP
How much VRAM does Carnice Qwen3.6 MoE 35B A3B APEX MTP need?
About 24.7 GB at its main file with an 8K context window. That is the 23.5 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 Carnice Qwen3.6 MoE 35B A3B APEX MTP on an 8 GB GPU?
Not at an 8K context. The smallest file needs about 12.9 GB, so you would need a larger GPU or CPU offloading.
What is the file size of Carnice Qwen3.6 MoE 35B A3B APEX MTP GGUF?
It depends on the quant: file is 11.7 GB, file is 14.3 GB, file is 17.3 GB, file is 17.3 GB, file is 23.5 GB, file is 23.5 GB and 3 more.
What context length does Carnice Qwen3.6 MoE 35B A3B APEX MTP 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 2.8 GB at 32K.
Where can I download Carnice Qwen3.6 MoE 35B A3B APEX MTP GGUF?
The files are on Hugging Face in the mudler/Carnice-Qwen3.6-MoE-35B-A3B-APEX-MTP-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 Carnice Qwen3.6 MoE 35B A3B APEX MTP 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 this model on 8 GB VRAM?
The smallest GGUF is 11.69 GB, which requires about 12.9 GB of VRAM at 8 k context. 8 GB VRAM is insufficient.
Does this model support vision?
No, the model has no vision capabilities.
What llama.cpp version is needed?
A recent llama.cpp (commit ≥255582687) is required to use the bundled MTP head.