google/gemma-4-26B-A4B-it

Gemma 4: Multimodal MoE for Advanced Reasoning and Coding

Google DeepMind's Gemma 4 family introduces multimodal models capable of processing text and image inputs (with audio on smaller variants) to generate text outputs. The google/gemma-4-26B-A4B-it is a 25.2 billion parameter instruction-tuned Mixture-of-Experts (MoE) model, featuring 3.8 billion active parameters for fast inference, making it suitable for consumer GPUs and workstations.

Key Capabilities & Advancements

• Multimodality: Handles Text, Image (variable aspect ratio/resolution), and Video inputs, with native audio support on E2B/E4B models.
• Reasoning: Designed with configurable thinking modes for highly capable reasoning.
• Extended Context Window: Supports a 256K token context window.
• Enhanced Coding & Agentic Capabilities: Achieves significant improvements in coding benchmarks and includes native function-calling support.
• Native System Prompt Support: Enables more structured and controllable conversations.
• Multilingual: Pre-trained on over 140 languages, with out-of-the-box support for 35+ languages.

Performance Highlights

The 26B A4B MoE model demonstrates strong performance across various benchmarks, including:

• MMLU Pro: 82.6%
• AIME 2026 no tools: 88.3%
• LiveCodeBench v6: 77.1%
• GPQA Diamond: 82.3%
• MMMU Pro (Vision): 73.8%

Ideal Use Cases

This model is well-suited for:

• Content Creation: Generating creative text formats, marketing copy, and email drafts.
• Conversational AI: Powering chatbots, virtual assistants, and interactive applications.
• Coding: Code generation, completion, and correction.
• Multimodal Understanding: Object detection, document/PDF parsing, UI understanding, chart comprehension, and OCR.
• Agentic Workflows: Leveraging native function-calling for structured tool use.