callmegrr/gemma-4-31B-it-heretic-z

VISIONConcurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 7, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The callmegrr/gemma-4-31B-it-heretic-z is a 31 billion parameter instruction-tuned multimodal language model, based on Google DeepMind's Gemma 4 architecture, with a 32768 token context window. This version is a decensored variant created using Heretic v1.4.0, offering enhanced reasoning, coding, and multimodal understanding capabilities for text, image, and video inputs. It is optimized for deployment on consumer GPUs and workstations, excelling in complex, long-context tasks.

Loading preview...

Model Overview

This model, callmegrr/gemma-4-31B-it-heretic-z, is a 31 billion parameter instruction-tuned variant of Google DeepMind's Gemma 4 model, specifically decensored using Heretic v1.4.0. It features a substantial 32768 token context window and is designed for advanced multimodal interactions, processing text, image, and video inputs to generate text outputs. The Gemma 4 architecture emphasizes strong reasoning, coding, and agentic capabilities, with native function-calling support and an enhanced system prompt role.

Key Capabilities

  • Multimodal Understanding: Processes text, images, and video (with variable aspect ratio and resolution) for comprehensive input interpretation.
  • Advanced Reasoning: Incorporates a built-in reasoning mode, allowing the model to perform step-by-step thinking before generating responses.
  • Extended Context: Supports a 256K token context window, enabling the handling of complex, long-context tasks.
  • Enhanced Coding: Demonstrates significant improvements in coding benchmarks and includes native function-calling for agentic workflows.
  • Decensored: This specific version is modified to be a decensored variant, offering different response characteristics compared to the original google/gemma-4-31B-it.

Good for

  • Complex Multimodal Applications: Ideal for scenarios requiring the integration of text, image, and video data.
  • Reasoning-Intensive Tasks: Suitable for applications benefiting from structured, step-by-step problem-solving.
  • Code Generation and Agentic Workflows: Excels in coding tasks and developing autonomous agents with tool-use capabilities.
  • Research and Experimentation: Provides a powerful base for exploring advanced multimodal LLM capabilities, particularly in contexts where decensored responses are desired.

Finetunes
1 models