---
base_model: lordx64/Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-Distilled
library_name: gguf
pipeline_tag: text-generation
tags:
- gguf
- llama.cpp
- lmstudio
- reasoning
- chain-of-thought
- qwen
- qwen3.6
- moe
- distillation
quantized_by: lordx64
license: apache-2.0
---

# Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-Distilled-GGUF

GGUF quantizations of [`lordx64/Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-Distilled`](https://huggingface.co/lordx64/Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-Distilled) for
use with [llama.cpp](https://github.com/ggerganov/llama.cpp) and
[LM Studio](https://lmstudio.ai/).

The base model is a reasoning-distilled variant of Qwen3.6-35B-A3B fine-tuned
to imitate the chain-of-thought style of Claude Opus 4.7. It thinks in explicit
`<think>...</think>` blocks before producing the final answer.

## Quant files

See the file list for all available quant levels. Common choices:

| File | Quant | Approx size | Use case |
|---|---|---|---|
| `*.IQ4_XS.gguf` | IQ4_XS | ~18 GB | Smallest quant with good quality — default pick for LM Studio |
| `*.Q4_K_M.gguf` | Q4_K_M | ~21 GB | Balanced quality / size |
| `*.Q5_K_M.gguf` | Q5_K_M | ~25 GB | Higher quality |
| `*.Q8_0.gguf`   | Q8_0   | ~35 GB | Near-lossless |

## Running in llama.cpp

```bash
llama-server \
  -m Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-Distilled.IQ4_XS.gguf \
  --host 127.0.0.1 --port 18081 \
  -c 32768 -fa on \
  --cache-type-k q8_0 --cache-type-v turbo4
```

## Running in LM Studio

Search for `lordx64/Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-Distilled-GGUF` inside LM Studio's model browser and pick the quant
that fits your RAM/VRAM. The model should appear automatically once HF indexes
this repo.

## License

Apache 2.0, inherited from the base model. See
[`lordx64/Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-Distilled`](https://huggingface.co/lordx64/Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-Distilled) for training details,
evaluations, and intended use.