---

base_model:

• schneewolflabs/Wichtelchen-Qwen3.5-9B

library_name: llama.cpp

license: apache-2.0

pipeline_tag: image-text-to-text

language:

• en

tags:

• gguf
• agents
• tool-use
• code
• hemlock
• qwen3.5

---

Wichtelchen-Qwen3.5-9B — GGUF

!image/png

GGUF quants of schneewolflabs/Wichtelchen-Qwen3.5-9B —

the Wichtel operator recipe on Qwen3.5-9B: delegation 10/10, Hemlock 56.1% on hembench.

See the main card for the full ladder and limitations.

| file | quant | use |

|---|---|---|

| Wichtelchen-Qwen3.5-9B-Q8_0.gguf | Q8_0 | reference quality — all benchmark numbers were measured on this |

| Wichtelchen-Qwen3.5-9B-Q6_K.gguf | Q6_K | near-Q8 quality, smaller |

| Wichtelchen-Qwen3.5-9B-Q5_K_M.gguf | Q5_K_M | balanced |

| Wichtelchen-Qwen3.5-9B-Q4_K_M.gguf | Q4_K_M | smallest recommended |

Serving

llama-server -m Wichtelchen-Qwen3.5-9B-Q8_0.gguf -ngl 99 -c 8192 --jinja -fa on -np 1 \
    --spec-type draft-mtp --spec-draft-n-max 4

The checkpoint carries the MTP head, so --spec-type draft-mtp speculative decoding works.

For vision, add the mmproj: --mmproj Wichtelchen-Qwen3.5-9B-mmproj-f16.gguf.