unsloth/gemma-4-12b-it

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 29, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The unsloth/gemma-4-12b-it is a 12 billion parameter instruction-tuned multimodal language model developed by Google DeepMind, part of the Gemma 4 family. This model uniquely integrates text, image, and audio inputs directly into a single decoder-only transformer, eliminating separate encoders for streamlined local execution. With a 256K token context window, it excels in reasoning, coding, and multimodal understanding tasks, making it suitable for consumer devices and workstations.

Loading preview...

Overview

unsloth/gemma-4-12b-it is a 12 billion parameter instruction-tuned model from the Gemma 4 family, developed by Google DeepMind. It stands out for its "Unified" encoder-free architecture, directly processing raw image patches and audio waveforms into the LLM's embedding space. This design reduces multimodal latency and allows for end-to-end fine-tuning. The model supports a substantial 256K token context window and is designed for efficient local execution.

Key Capabilities

  • Multimodal Understanding: Natively processes text, image, and audio inputs without separate encoders.
  • Reasoning: Features configurable thinking modes for step-by-step problem-solving.
  • Long Context: Supports a 256K token context window for complex tasks.
  • Coding & Agentic Capabilities: Enhanced performance in coding benchmarks and native function-calling support.
  • Multilingual: Pre-trained on over 140 languages with out-of-the-box support for 35+.
  • Optimized for On-Device: Designed for efficient deployment on consumer GPUs and workstations.

Good for

  • Multimodal Applications: Ideal for scenarios requiring combined text, image, and audio processing.
  • Reasoning Tasks: Benefits from its built-in reasoning mode for logical problem-solving.
  • Code Generation: Strong performance in coding tasks, including generation, completion, and correction.
  • Agentic Workflows: Supports structured tool use for autonomous agents.
  • Local Deployment: Optimized for efficient execution on consumer hardware.