LLMPrivate

Qwen3.6 35B A3B Uncensored

An uncensored, highly optimized 35B Mixture-of-Experts model from the Qwen 3.6 family, built for agentic coding and unrestricted reasoning.

Maker
Alibaba Cloud (base) / HauhauCS (uncensored)
Modality
Text
Context
128K tokens
Open weights
Yes (Apache-2.0 base)

Overview

What is Qwen3.6 35B A3B Uncensored

Qwen3.6 35B A3B Uncensored is an unrestricted, open-weight Mixture-of-Experts (MoE) model released in April 2026. Fine-tuned to eliminate safety refusals, it excels at agentic coding, repository-level reasoning, and complex instruction-following without preachy filters or moralizing interruptions.

Running it privately on Venice

On Venice, you can run Qwen3.6 35B A3B Uncensored with absolute sovereignty. Your prompts are fully private, protected by end-to-end encryption and processed in a secure Trusted Execution Environment (TEE) with zero data retention. This ensures your uncensored creative writing, coding, and research remain entirely yours.

Private (zero retention)No prompt trainingTEE · hardware enclaveEnd-to-end encrypted

Assessment

Strengths and limitations

Strengths
  • 100% refusal-free generation, making it ideal for creative writing, roleplay, and complex technical tasks that trigger standard safety filters.
  • Superb agentic coding capabilities, handling frontend workflows and repository-level reasoning with high precision.
  • Highly efficient Mixture of Experts (MoE) architecture, offering fast inference speeds by only activating necessary parameters.
  • Integrated web search capability on Venice, allowing the model to ground its uncensored responses in real-time web data.
Limitations
  • Complete lack of safety filters means it can generate toxic, biased, or highly sensitive content if prompted to do so.
  • 128K context window is highly capable but smaller than massive proprietary models like Claude's 1M token window.
  • Text-only modality with no native support for image or audio inputs.

Capabilities

What it supports

  • Tool use / function calling
  • Vision (image input)
  • Reasoning
  • Web search
  • Code-optimized
  • Structured output (JSON schema)
  • Audio input
  • Video input
  • Multiple image inputs
  • Log probabilities

Specifications

Datasheet

Maker
Alibaba Cloud (base) / HauhauCS (uncensored)
Released
April 16, 2026
Architecture
Mixture of Experts (MoE) with Gated DeltaNet
Active Parameters
8 Routed + 1 Shared Expert
Total Parameters
35B
Open weights
Yes (Apache-2.0 base)
Context window
128K tokens
Max output
4.096K tokens
Capabilities
Web search
Privacy on Venice
Private — zero retention
Available on Venice since
May 2026

API

Call it from your code

Venice exposes an OpenAI-compatible API. Point your base URL at Venice and pass the model id.

curl https://api.venice.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "e2ee-qwen3-6-35b-a3b-uncensored-p",
    "messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
  }'

Pricing

What it costs on Venice

Billed per token on Venice: $0.38 per 1M input tokens and $1.88 per 1M output tokens.

Input / 1M tokens
$0.38
Per 1M tokens
Output / 1M tokens
$1.88
Per 1M tokens

New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.

Alternatives

How it compares

ModelContext windowStrongest atOpen weightsPrice (Venice)
Qwen3.6 35B A3B Uncensored128K tokensUncensored coding & reasoningNo$0.38 / $1.88
DeepSeek V3.2160K tokensDeep reasoning & mathYes$0.33 / $0.48
Google Gemma 4 31B Instruct256K tokensInstruction followingYes$0.12 / $0.36
Claude Sonnet 4.61M tokensGeneral intelligenceNo$3.60 / $18

The go-to choice for private, unrestricted coding and reasoning.

Use cases

What it is good for

  1. 01Unrestricted creative writing, fiction, and complex character roleplay.
  2. 02Advanced software engineering, debugging, and multi-file repository analysis.
  3. 03Private research on sensitive, controversial, or highly regulated topics without AI lecturing.
  4. 04Real-time web-grounded research and coding assistance.

Prompting

Getting better results

Be direct and explicit; because the model is uncensored, you do not need to use complex 'jailbreak' framing.

Use the web search capability for coding queries involving recently updated libraries or APIs.

Provide clear system instructions to guide the tone, as uncensored models can easily drift in style.

Version history

Qwen 3.5 35B A3B
2026-02

Previous generation MoE model.

Qwen 3.6 35B A3B
2026-04-16

Base open-weight release with improved agentic coding.

Qwen3.6 35B A3B Uncensored
2026-04-22

Uncensored fine-tune by HauhauCS.

FAQ

Frequently asked questions

Qwen3.6 35B A3B Uncensored is a 35-billion parameter Mixture-of-Experts (MoE) model from Alibaba's Qwen 3.6 family, modified by the developer HauhauCS to completely remove safety refusals and guardrails.

On Venice, it is billed per token at $0.38 per 1 million input tokens and $1.88 per 1 million output tokens.

Yes, the base model is open-weight and released under the Apache 2.0 license. The uncensored fine-tune weights are also openly shared on Hugging Face.

Yes, on Venice, this model is equipped with real-time web search capabilities, allowing it to fetch up-to-date information for your queries.

While DeepSeek V3.2 is an exceptional reasoning model, it contains standard safety alignments. Qwen3.6 35B Uncensored offers comparable technical capabilities in coding and reasoning but with zero refusals.

A3B refers to the specific architectural design of this Qwen variant, which utilizes a Mixture-of-Experts setup with 8 routed experts and 1 shared expert to optimize active parameter efficiency.

Absolutely. Venice runs this model with end-to-end encryption in a Trusted Execution Environment (TEE) under a strict zero-retention policy, meaning your prompts are never stored, inspected, or used for training.

Run Qwen3.6 35B A3B Uncensored privately

No prompt logging. No data used for training.