xtools-at/Qwen3.5-4B-MultiDistill-Megamix-8xMerge-UNCENSORED-heretic

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

xtools-at/Qwen3.5-4B-MultiDistill-Megamix-8xMerge-UNCENSORED-heretic is a 4.5 billion parameter language model based on the Qwen3.5-4B architecture, created by xtools-at. This model is a decensored version of its predecessor, achieved using the Heretic v2.0.0.dev0 tool, and is a linear weighted merge of three Qwen3.5-4B-based checkpoints focused on reasoning distillation, Claude-derived training data, and Heretic-derived variants. It is designed for general-purpose conversation, instruction following, mathematical and logical problem-solving, and code-related tasks, with a context length of 32768 tokens.

Loading preview...

Overview of Qwen3.5-4B-MultiDistill-Megamix-8xMerge-UNCENSORED-heretic

This model, developed by xtools-at, is a 4.5 billion parameter language model built upon the Qwen3.5-4B architecture. It represents a significant modification of the original xtools-at/Qwen3.5-4B-MultiDistill-Megamix-8xMerge by being decensored using the Heretic v2.0.0.dev0 tool. The model is a linear weighted merge of three distinct Qwen3.5-4B checkpoints, each contributing specialized capabilities: one focused on Claude-derived reasoning, another on Heretic-related lineage, and a third on multi-distillation techniques. This combination aims to integrate complementary behaviors while preserving the base Qwen3.5-4B's general text generation strengths.

Key Capabilities

  • Decensored Output: Significantly reduced refusal rates (3/100 compared to 95/100 in the original model), offering less restricted responses.
  • Enhanced Reasoning: Incorporates lineages from Claude-derived reasoning data and Qwen3.5-generated reasoning examples, supporting mathematical, logical, and analytical problem-solving.
  • Code and Technical Explanations: Benefits from source models trained with data relevant to code-related questions and technical discussions.
  • Structured Responses: Designed to handle multi-step tasks and generate structured outputs.
  • Agent-Oriented Tasks: Potential for use in agent-style or tool-use workflows, inferred from source model lineages and dataset provenance.

Good For

  • Experimentation with Decensored Models: Ideal for research into the effects of decensoring on model behavior and utility.
  • General-Purpose AI Assistants: Suitable for conversational agents and instruction-following applications requiring broad capabilities.
  • Complex Problem Solving: Useful for tasks demanding strong mathematical, logical, or analytical reasoning.
  • Developer Tools: Can assist with code generation, debugging, and providing technical explanations.
  • Investigating Model Merging: Provides a case study for combining diverse fine-tuning objectives (reasoning, agentic behavior) into a single compact model.

Finetunes
1 models