google/gemma-4-31B-it-qat-q4_0-unquantized

VISIONConcurrency Cost:2Model Size:31BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Apr 28, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

The google/gemma-4-31B-it-qat-q4_0-unquantized model is a 31 billion parameter instruction-tuned multimodal language model from the Gemma 4 family, developed by Google DeepMind. This unquantized QAT checkpoint is optimized for custom downstream compilation and research, offering a 256K token context window. It excels in reasoning, coding, and multimodal understanding, processing text and image inputs to generate text outputs, with strong performance across various benchmarks.

Loading preview...

Gemma 4 31B Instruction-Tuned QAT (Unquantized) Overview

This model is part of the Gemma 4 family by Google DeepMind, featuring 31 billion parameters and a substantial 256K token context window. It is an unquantized checkpoint from a Quantization-Aware Training (QAT) pipeline, designed to maintain high quality while enabling memory-efficient deployment. The Gemma 4 models are multimodal, capable of processing text and image inputs (with audio support on smaller variants) to generate text outputs, and offer multilingual support across over 140 languages.

Key Capabilities

  • Advanced Reasoning: Designed with configurable thinking modes for highly capable reasoning.
  • Extended Multimodalities: Processes text and images with variable aspect ratio and resolution support. Video input is also supported by processing sequences of frames.
  • Enhanced Coding & Agentic Capabilities: Achieves significant improvements in coding benchmarks and includes native function-calling for autonomous agents.
  • Increased Context Window: Supports a 256K token context window, enabling complex, long-context tasks.
  • Native System Prompt Support: Introduces native support for the system role for more structured conversations.

Good For

  • Custom Downstream Compilation & Research: Ideal for developers and researchers requiring high-quality, unquantized weights for specialized applications.
  • Reasoning and Agentic Workflows: Excels in tasks requiring logical deduction and autonomous agent development.
  • Coding Tasks: Strong performance in code generation, completion, and correction.
  • Multimodal Understanding: Suitable for applications involving interleaved text and image inputs, such as object detection, document parsing, and chart comprehension.

Finetunes
2 models