Jinx-org/Jinx-Qwen3-32B

Jinx-Qwen3-32B: An Unfiltered LLM for AI Safety Research

Jinx-org/Jinx-Qwen3-32B is a 32 billion parameter language model derived from the Qwen3 architecture, uniquely engineered for AI safety research. Unlike typical LLMs, Jinx is a "helpful-only" variant that provides responses to all queries without applying safety filters or refusals.

Key Characteristics

• Zero Refusal Rate: Designed to respond to every query, including those that might typically trigger safety filters in other models.
• Preserved Capabilities: Maintains strong reasoning and instruction-following abilities, comparable to its base Qwen3 model, ensuring research relevance.
• Research-Focused: Exclusively intended for studying alignment failures, evaluating safety boundaries, and understanding the behavior of LLMs without inherent safety guardrails.

Important Usage Advisory

Due to its unfiltered nature, Jinx-Qwen3-32B carries significant risks and is intended for a restricted audience:

• Unfiltered Content Risk: May produce offensive, controversial, or socially sensitive material. All outputs require thorough human verification.
• Restricted Audience: Unsuitable for minors, public deployments, or high-risk applications (e.g., medical, legal, financial).
• User Accountability: Users assume full liability for compliance with laws, ethical implications, and any damages from generated content.

This model is a critical tool for researchers investigating the limits and behaviors of LLMs in an unfiltered environment, providing insights into potential alignment issues.