ark4004/gemma-4-31B-it-jh

VISIONConcurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 12, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Gemma 4 31B is a 30.7 billion parameter multimodal language model developed by Google DeepMind, part of the Gemma 4 family. This instruction-tuned variant processes text and image inputs, generating text outputs, and features a 256K token context window. It is designed for advanced reasoning, coding, and agentic workflows, offering strong performance across various benchmarks including MMLU Pro (85.2%) and LiveCodeBench v6 (80.0%).

Loading preview...

Gemma 4 31B: Advanced Multimodal Reasoning and Coding

This model is the 31 billion parameter dense variant from Google DeepMind's Gemma 4 family, an instruction-tuned multimodal LLM. It excels in processing both text and image inputs, generating text outputs, and supports a substantial 256K token context window. Gemma 4 models are built with configurable thinking modes for enhanced reasoning and feature native function-calling support, making them suitable for complex agentic workflows.

Key Capabilities

  • Multimodal Understanding: Processes text and images with variable aspect ratio and resolution support. While other Gemma 4 models support audio, the 31B variant focuses on text and image.
  • Advanced Reasoning: Designed with built-in reasoning modes for step-by-step thought processes before generating answers.
  • Extended Context: Features a 256K token context window, enabling handling of long and complex inputs.
  • Enhanced Coding: Demonstrates significant improvements in coding benchmarks (e.g., 80.0% on LiveCodeBench v6) and supports code generation, completion, and correction.
  • Agentic Workflows: Includes native function-calling support for structured tool use, facilitating the development of autonomous agents.
  • Multilingual Support: Pre-trained on over 140 languages with out-of-the-box support for 35+ languages.

Performance Highlights

The Gemma 4 31B model achieves strong benchmark results, including 85.2% on MMLU Pro, 89.2% on AIME 2026 (no tools), and 84.3% on GPQA Diamond. For vision tasks, it scores 76.9% on MMMU Pro and 85.6% on MATH-Vision. Its architecture employs a hybrid attention mechanism combining local sliding window attention with global attention for efficient long-context processing.

Ideal Use Cases

This model is well-suited for applications requiring high-performance reasoning, complex code generation, multimodal content analysis (text and image), and the development of sophisticated AI agents. Its large context window and robust capabilities make it a strong candidate for demanding enterprise and research applications.