Gemma 4

What is Gemma 4?

Gemma 4 is a versatile AI model family developed by Google, designed for a variety of reasoning and multimodal tasks. It comes in multiple configurations, including E2B, E4B, 12B, 26B A4B, and 31B variants, allowing users the flexibility to choose a model that suits their specific needs regarding quality, latency, and hardware costs. This breadth of model sizes caters to different environments, whether local or hosted, enhancing deployment options significantly.

"Gemma 4 is easier to evaluate because the official family now covers edge-friendly sizes, a unified 12B multimodal model, and a dense 31B model for quality-first workloads."

Gemma 4 supports a wide range of input types, with some models accepting image, video, and audio input. Models like E2B and E4B feature context support from 128K to 256K, making them ideal for complex, long-document analyses and agent workflows. Additionally, it employs a highly permissive Apache 2.0 license, allowing for self-hosting and customization advantages.

Features

• Five official sizes: E2B, E4B, 12B, 26B A4B, and 31B for varying quality and performance.
• High context capacity: Supports 128K to 256K context for extensive document handling.
• Multimodal by default: Accepts text, image, video, and audio inputs depending on the model.
• Flexible deployment: Can be run locally using platforms like LM Studio and Ollama, or through hosted variants.
• Memory guidance: Approximately ranging from 2.9 GB for the E2B model to 17.5 GB for the 31B model.

Gemma 4 provides a robust environment for developers looking for a flexible and powerful AI tool for various applications, fully accommodating different hardware setups and user requirements.