darkc0de/gemma-4-31B-it-Claude-Opus-Distill-v2-heretic

Hugging Face
VISIONConcurrency Cost:2Model Size:31BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 8, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

The darkc0de/gemma-4-31B-it-Claude-Opus-Distill-v2-heretic is a 31 billion parameter Gemma 4-based instruction-tuned language model, developed by darkc0de, that has been decensored from the original TeichAI/gemma-4-31B-it-Claude-Opus-Distill-v2. This model is fine-tuned to absorb high-effort reasoning distillation from Claude-4.6 Opus interactions, making it highly optimized for complex problem-solving in coding, science, and deep research. It supports a 32768 token context length and excels at delivering precise, nuanced solutions across demanding domains.

Loading preview...

Model Overview

This model, darkc0de/gemma-4-31B-it-Claude-Opus-Distill-v2-heretic, is a decensored version of the TeichAI/gemma-4-31B-it-Claude-Opus-Distill-v2 model, created using the Heretic v1.3.0 tool. It is built upon the Gemma 4 - 31B architecture and fine-tuned with high-effort reasoning datasets derived from Claude-4.6 Opus interactions, leveraging the Unsloth framework for efficiency. The decensoring process significantly reduces refusals, with this model showing 28/100 refusals compared to the original's 89/100.

Key Capabilities

  • Enhanced Reasoning: Absorbs state-of-the-art reasoning logic from Claude-4.6 Opus, excelling in complex problem-solving.
  • Decensored Output: Provides less restricted responses compared to its base model, as indicated by lower refusal rates.
  • Structured Thinking: Utilizes a structured and efficient thinking pattern, with optional thinking mode configuration.
  • Multimodal Support: While the base model supports text, audio, image, and video inputs, the provided code snippets demonstrate how to use AutoModelForMultimodalLM for image, audio, and video processing.

Good For

  • πŸ’» Coding: Advanced code generation, debugging, and software architecture planning.
  • πŸ”¬ Science: Deep scientific reasoning, hypothesis evaluation, and analytical problem-solving.
  • πŸ”Ž Deep Research: Navigating complex, multi-step research queries and synthesizing vast amounts of information.
  • 🧠 General Purpose: Highly capable instruction-following for everyday tasks requiring high logical coherence, especially where less restrictive output is desired.