Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated
This is a 35.9B parameter mixture-of-experts model derived from Qwen 3.6, fine-tuned with Claude 4.7 reasoning distillation and then modified ('abliterated') to remove safety filtering. The model card explicitly warns that safety guardrails have been significantly reduced and outputs may be sensitive, controversial, or inappropriate. It is positioned for research and controlled environments, not production or public-facing use.
Key facts
Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.
| Field | Value |
|---|---|
| Developer | huihui-ai |
| Parameters | 36B |
| Context window | Unknown |
| License | apache-2.0 — OSI-approved |
| Modality / task | text-generation |
| Gated on HuggingFace | No |
| Downloads | 53k |
| Likes | 159 |
| Last updated | 2026-04-21 |
| Source | huihui-ai/Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated |
What Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated is
A Qwen3.6-35B MoE variant with LoRA adapter applied, distilled from Claude 4.7 reasoning behavior, and processed with the remove-refusals-with-transformers abliteration technique to strip refusal mechanisms. Distributed in safetensors format; context length not specified. Last modified April 2026. Not gated; 53k downloads.
Run Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated locally
Load the open weights with 🤗 Transformers and generate — the same model, self-hosted.
from transformers import pipelinepipe = pipeline("text-generation", model="huihui-ai/Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated")out = pipe("Explain retrieval-augmented generation in one sentence.", max_new_tokens=128)print(out[0]["generated_text"])Swap in vLLM or Ollama for production-grade serving. DEV.co can stand up the inference stack.
How you'd run it
A typical self-hosted path — open weights, an inference server, your application.
DEV.co builds each layer — from GPU infrastructure to the application.
Best use cases
Running & fine-tuning it
ESTIMATE: ~72 GB VRAM for 35.9B full-precision FP32 inference; ~36 GB for FP16 (typical); MoE sparsity may reduce activation cost. Exact precision and sparsity details not specified in card—requires testing. Ollama and llama.cpp support quantization; verify on target hardware.
Model already incorporates LoRA adapter from lordx64 base; additional fine-tuning feasible via LoRA or QLoRA on consumer GPUs (e.g., 24–48 GB VRAM with quantization). Unsloth framework mentioned in tags; supports efficient fine-tuning. Abliteration is a post-hoc weight modification, not a training checkpoint—retraining safety features would require separate intervention.
When to avoid it — and what to weigh
- Production or Public-Facing Applications — The model card explicitly discourages production deployment or direct public exposure due to reduced safety optimization and risk of inappropriate outputs.
- Regulatory or Compliance-Heavy Domains — Avoid in regulated industries (finance, healthcare, government) where content moderation, audit trails, and safety guarantees are non-negotiable.
- Audience with Minors or Sensitivity Requirements — Not suitable for applications serving underage users, vulnerable populations, or contexts where content appropriateness is critical.
- High-Liability Use Cases — The model card disclaims responsibility for consequences; unsuitable where liability for harmful outputs falls on the deployer (legal risk acceptance required).
License & commercial use
Apache 2.0 license. This is a permissive, OSI-approved open-source license permitting commercial use, modification, and distribution with attribution and liability disclaimer.
Apache 2.0 permits commercial use, modification, and redistribution. However, exercise extreme caution: (1) The model card warns against production/public-facing deployment and disclaims responsibility for harmful outputs. (2) Liability for generated content rests entirely with the deployer. (3) Legal and ethical risk assessment in your jurisdiction is mandatory—particularly if the model generates sensitive, defamatory, or regulated content. (4) Commercial viability hinges on your ability to manage safety, compliance, and reputational risk independently. Consult legal counsel.
DEV.co evaluation signals
Editorial assessment — not user reviews. Directional, with an explicit confidence level.
| Signal | Assessment |
|---|---|
| Maintenance | Unknown |
| Documentation | Limited |
| License clarity | Clear |
| Deployment complexity | Moderate |
| DEV.co fit | Possible |
| Assessment confidence | Medium |
No explicit security audit or vulnerability disclosure process mentioned. Key considerations: (1) Abliteration removes safety mechanisms—outputs may violate content policies or generate harmful text. (2) No details on model poisoning, data provenance, or training data vetting. (3) LoRA adapter and abliteration process are third-party modifications; integrity/authenticity not formally verified. (4) Self-hosting eliminates API-level moderation but introduces infrastructure security burden. (5) Legal liability for generated content is entirely on deployer. Deploy only in controlled environments with manual output review and monitoring in place.
Alternatives to consider
Meta Llama 3.1 405B or 70B (standard checkpoint)
Production-ready, well-maintained, safety-aligned baseline with strong community support and clear commercial licensing (Llama Community License). Better for enterprise or public-facing use.
Mistral 7B / Mixtral 8x7B
Smaller, efficient alternatives with permissive Apache 2.0 license, active maintenance, and more mature safety tuning. Suitable for resource-constrained deployments.
OpenAI GPT-4 / Claude API (managed service)
If safety, compliance, and reliability are critical, managed APIs offload infrastructure and liability to vendor. Higher cost but lower operational risk.
Ship Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated with senior software developers
This model requires explicit risk acceptance and is best suited for controlled research environments. For production deployments, safety-critical applications, or commercial use with liability concerns, consult with our team to evaluate alternatives (Llama, Mistral) or managed APIs. We can help you balance capability, compliance, and operational complexity.
Talk to DEV.coRelated open-source tools
Surfaced by semantic similarity across the DEV.co open-source index.
Related on DEV.co
Explore the category and the services that help you build with it.
Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated FAQ
Can I use this model commercially?
How much GPU VRAM do I need?
What is 'abliteration' and why is it important?
Is this model maintained?
Work with a software development agency
Adopting Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated is usually one piece of a larger software development effort. As a software development agency, DEV.co provides software development services and web development expertise — pairing senior software developers and web developers with your team to design, build, and operate open-source llms software in production.
Evaluating This Model for Your Use Case?
This model requires explicit risk acceptance and is best suited for controlled research environments. For production deployments, safety-critical applications, or commercial use with liability concerns, consult with our team to evaluate alternatives (Llama, Mistral) or managed APIs. We can help you balance capability, compliance, and operational complexity.