richardyoung/Qwen2.5-14B-Instruct-heretic

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The richardyoung/Qwen2.5-14B-Instruct-heretic is a 14.8 billion parameter instruction-tuned causal language model, based on the Qwen2.5 architecture, with a context length of 32,768 tokens. This version is a 'decensored' variant of the original Qwen/Qwen2.5-14B-Instruct, created using the Heretic v1.4.0 tool. It offers significantly reduced refusals compared to the original model, while retaining Qwen2.5's enhanced capabilities in coding, mathematics, long text generation, and structured data understanding.

Loading preview...

Model Overview

This model, richardyoung/Qwen2.5-14B-Instruct-heretic, is a 14.8 billion parameter instruction-tuned causal language model derived from the Qwen2.5 series. It is a modified version of the original Qwen/Qwen2.5-14B-Instruct, specifically 'decensored' using the Heretic v1.4.0 tool. This modification significantly reduces the model's refusal rate, making it more permissive in its responses.

Key Characteristics & Differentiators

  • Decensored Behavior: Compared to the original Qwen2.5-14B-Instruct, this 'heretic' version exhibits a substantially lower refusal rate (11/100 vs. 98/100), indicating a more open and less restrictive response generation.
  • Enhanced Core Capabilities: Inherits the Qwen2.5 improvements, including greater knowledge, significantly improved capabilities in coding and mathematics, and better instruction following.
  • Long Context Support: Supports a full context length of 131,072 tokens (with YaRN scaling) and can generate up to 8,192 tokens, making it suitable for processing and generating extensive texts.
  • Structured Data & Output: Demonstrates improved understanding of structured data (e.g., tables) and enhanced ability to generate structured outputs, particularly JSON.
  • Multilingual: Offers robust multilingual support for over 29 languages, including major global languages like Chinese, English, French, Spanish, German, and Japanese.

Use Cases

This model is particularly well-suited for applications requiring a powerful, instruction-following LLM with a high degree of permissiveness. It excels in scenarios where the original Qwen2.5 might refuse certain prompts due to safety alignments, making it ideal for research into model behavior, creative content generation without strict guardrails, or applications where a broader range of responses is desired. Its strong coding, mathematical, and long-context capabilities further broaden its utility for complex technical and analytical tasks.

Finetunes
1 models