# Gemma

- Category: Models & resources
- Official site: https://ai.google.dev/gemma?hl=zh-cn
- HTML: https://www.88z.ai/en/tools/gemma
- Markdown: https://www.88z.ai/en/tools/gemma.md

Gemma is a family of open-weight AI models built by Google DeepMind, designed to help developers build responsible and efficient AI applications at scale.

## What is Gemma?

Gemma is a family of open-weight AI models built by Google DeepMind, designed to help developers build responsible and efficient AI applications at scale.As the core of the 'Open models' lineup, Gemma offers variants ranging from lightweight edge deployments to high-performance desktop inference, ensuring intelligence runs wherever users need it.

The latest Gemma 4 series is described by Google as the 'most capable open models byte for byte,' purpose-built for advanced reasoning and agentic workflows.The family includes E2B/E4B versions optimized for mobile and IoT efficiency, alongside 12B, 26B, and 31B parameter models that deliver frontier-level intelligence on personal computers.

Beyond base language models, the Gemma ecosystem extends into multimodal and specialized domains.With dedicated variants like EmbeddingGemma, DiffusionGemma, MedGemma, and ShieldGemma, developers can leverage unified embedding spaces or specific capabilities for text, image, audio, and medical contexts to build richer and safer applications.

## Key features

- **Ubiquitous Deplo**：Offers model sizes for cloud servers, personal computers, phones, and IoT devices, enabling true run-anywhere capability.
- **Gemma 4 Architec**：The latest generation is purpose-built for advanced reasoning and agentic workflows, balancing parameter count with intelligence.
- **Multimodal Unity**：Models like EmbeddingGemma 2 unify code, images, video, and audio in a shared embedding space.
- **Domain Specializ**：Includes MedGemma for medical imaging interpretation and TranslateGemma for communication across 55 languages.
- **Safety & Complia**：ShieldGemma 2 provides modular classifiers for detecting policy-violating content; VaultGemma offers differential privacy.
- **Efficient Infere**：Enhanced speed on mobile and laptops via Quantization-Aware Training (QAT) and multi-token prediction drafters.

## How to use it

1. Visit Google AI Studio or the official Gemma website to browse available model variants and technical specifications.
2. Select the appropriate Gemma model size and type based on your target hardware (mobile, laptop, or cloud).
3. Consult the official documentation for integration guides tailored to frameworks like JAX, PyTorch, or TensorFlow.
4. Use the provided toolchains to fine-tune or quantize the model for optimal performance in your specific use case.
5. Deploy the model locally or in the cloud, leveraging tools like ShieldGemma to ensure application safety.

## Where it fits

- **Mobile AI Assist**：Run low-latency, high-privacy local AI features on phones or IoT devices using efficient models like E2B/E4B.
- **Medical Data Ana**：Use MedGemma to process high-dimensional medical images and text, assisting professionals in analysis and interpretation.
- **Cross-Language C**：Enable high-quality translation across 55 languages with TranslateGemma to break down global collaboration barriers.
- **Content Safety M**：Integrate ShieldGemma 2 to automatically detect and filter policy-violating content, ensuring platform safety and compliance.

## Pricing and free quota

Gemma consists of open-weight models; specific costs depend on deployment method (local or cloud).Pricing follows the official site.

## Who it is for

- AI Application Developers
- Mobile & Embedded Engineers
- Healthcare Tech Researchers
- Enterprise Safety & Compliance Teams

## FAQ

### Are Gemma models free to use?

Gemma consists of open-weight models that developers can download and run on their own infrastructure.Please refer to the official license for specific terms and usage restrictions.

### What are the main improvements in Gemma 4?

Gemma 4 is described as the most capable open models byte for byte, purpose-built for advanced reasoning and agentic workflows, with improved compression and multimodal capabilities.

### How can I run Gemma on mobile devices?

You can choose variants like E2B or E4B designed for maximum compute and memory efficiency, and optimize them using techniques like Quantization-Aware Training (QAT).

### Which programming languages or frameworks does Gemma support?

Official documentation typically supports major deep learning frameworks.Please check Google AI Studio or the official docs for the latest integration guides.

## Tags

Open Weights · Multimodal AI · Edge Computing · Google DeepMind

Features, pricing and availability follow the official site. Except for 88z.ai’s own API pages, this is a third-party product; 88z.ai only provides a directory introduction.