Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated — Q4_K_M GGUF
- This model has been tested by myself in Hermes Agent, it does everything it is supposed to, text, coding, vision, thinking, tools usage, everything works perfectly well so far there. I have also found that this very specific Opus 4.6 Distillation does not use as many emojis as the non-Opus one, which is wave:great.
This is a Q4_K_M GGUF quantization of huihui-ai/Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated.
Refer to the original model card for full details, usage warnings, and licensing information.
This is a Qwen3.5-35B-A3B MoE model that has been:
- Fine-tuned in Claude 4.6 Opus style — targets Claude Opus-level instruction following, reasoning depth, and structured output quality
- Abliterated — refusal mechanisms removed; no alignment restrictions
- Vision-capable — the original weights include a full vision encoder (
preprocessor_config.json,video_preprocessor_config.json); an mmproj file can be generated from the source safetensors for multimodal use
Files
| File | Size | Description |
|---|---|---|
Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M.gguf | ~20 GB | Main text model (Q4_K_M) |
Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-mmproj-F16.gguf | ~858 MB | Vision projector for image input |
Details
| Source model | huihui-ai/Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated |
| Architecture | qwen35moe (35B total params, ~3B active; 256 experts, 8 active per token) |
| Quantization | Q4_K_M (~4.88 BPW) |
| File size | ~20 GB + 858 MB mmproj |
| Quantized with | llama.cpp b8352 |
| Context length | 262,144 tokens (trained) |
Usage with llama.cpp
llama-cli \
--hf-repo cesarsal1nas/Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M-GGUF \
--hf-file Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M.gguf \
-p "Tell me about the universe"
llama-server \
--hf-repo cesarsal1nas/Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M-GGUF \
--hf-file Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M.gguf \
-c 131072
With vision (image input)
llama-server \
--hf-repo cesarsal1nas/Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M-GGUF \
--hf-file Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M.gguf \
--mmproj Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-mmproj-F16.gguf \
-c 131072
Usage with Ollama
ollama run hf.co/cesarsal1nas/Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated-Q4_K_M-GGUF
Credits
- Fine-tune + abliteration by huihui-ai — Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliterated
- Base model by Qwen — Qwen3.5-35B-A3B
- Hermes Agent (as mentioned in the model card) by NousResearch