HauhauCS
Qwen3.6 35B-A3B Aggressive
A highly downloaded community release focused on stronger refusal reduction.
- License
- Apache 2.0
Hosted on VynarisQwen3.6
Qwen3.6 35B-A3B Uncensored
The fast hosted Vynaris option for high-volume authorized security testing, agents, and reduced-refusal research.
- Scale
- 35B total / about 3B active class
- Architecture
- Sparse mixture-of-experts
- Modality
- Text API; upstream family supports images
- Context
- 128K through the Vynaris API
Model card
What is Qwen3.6 35B-A3B Uncensored?
Qwen3.6 35B-A3B uncensored variants adapt Qwen's sparse 35B family for fewer refusals. Although the model has a large total expert pool, it activates only a small fraction for each token, making it attractive for responsive serving.
Vynaris exposes a stable text-generation profile through an OpenAI-compatible API. Users who prefer self-management can compare community releases and follow each publisher's runtime instructions.
Capabilities
Where this family fits
- High-volume adversarial prompt and policy testing
- Authorized security agents that need low latency
- Defensive code review and threat-model iteration
- Private research where standard refusal behavior obstructs evaluation
Limitations
What to verify first
- Prioritizes throughput over the depth of larger dense or frontier-scale options
- The Vynaris endpoint currently focuses on text generation
- Reduced-refusal tuning can make answers less calibrated or cautious
Community ecosystem
Qwen3.6 35B-A3B Uncensored variants on Hugging Face
These are independent community releases. Vynaris is not affiliated with their publishers and does not imply that similarly named variants are identical. Check each repository card and license before use.
AEON-7
Qwen3.6 35B-A3B Heretic
A community heretic adaptation of the same sparse Qwen family.
- License
- Apache 2.0
Where to use it
Hosted, community, and self-managed paths
Vynaris hosted API
Use one API key and select this model in each OpenAI-compatible request. No model infrastructure to operate.
Get an API key →Hugging Face repositories
Inspect model cards, licenses, files, release notes, and community discussions directly from the publishers listed above.
Open the first listed variant ↗Local or private deployment
Follow the selected repository's documented runtime path. Hardware, context, batching, and compatibility depend on that exact distribution.
Read Vynaris API documentation →Hosted API
vynaris/qwen3.6-35b-a3b-uncensored
Put this exact ID in the standard model field. The first request after an idle period can take longer while private capacity starts.
curl https://api.vynaris.com/v1/chat/completions \
-H "Authorization: Bearer $VYNARIS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "vynaris/qwen3.6-35b-a3b-uncensored",
"messages": [{"role": "user", "content": "Review this authorized security test..."}],
"stream": true
}'Price and access
$1.00 input / $5.00 output
Prices are per million tokens. A successful purchase of at least $50 in API credit qualifies the account; access remains available as the balance is spent down.
Data handling
No prompt or output retention
Vynaris does not persist request or response bodies for this model ID. Billing and operational metadata such as model, token counts, cost, latency, account, status, and request ID is retained.
FAQ
Questions about Qwen3.6 35B-A3B Uncensored
Is Qwen3.6 35B-A3B uncensored hosted by Vynaris?
Yes. Use vynaris/qwen3.6-35b-a3b-uncensored in the model field of an OpenAI-compatible chat-completions request.
What does A3B mean?
It indicates that roughly 3B parameters are active for each token even though the total expert pool is much larger. Memory and throughput still depend on the full serving system.
How much context does Vynaris support?
The current hosted API profile supports up to 128K tokens. Long contexts can materially affect latency and capacity.
Are prompts and outputs stored?
Prompt and output bodies are not persisted for this hosted model ID. Vynaris retains billing and operational metadata needed to run the service.
Related model cards
Compare nearby size and architecture classes
Security-friendly, not criminal-friendly.
Use these models only for lawful, authorized work. Child sexual exploitation, human trafficking or exploitation, and non-consensual sexual content are prohibited. Reduced refusal is not a safety, accuracy, or authorization guarantee.