Key Highlights

• Google releases Gemma 4 under the Apache 2.0 open-source license
• The model supports reasoning, coding assistance, audio, and video understanding
• It runs locally on devices, including some laptops and Android hardware
• Available in multiple sizes from 2B to 31B parameters

Google has launched Gemma 4, its latest open large language model, and made it fully open source under the Apache 2.0 license. The release allows developers to download, modify, and deploy the model locally without subscription costs. With reasoning upgrades, multimodal support, and flexible deployment options, Gemma 4 marks a shift in how Google positions open AI models.

What is Gemma 4 and why is it significant?

Gemma 4 is Google’s newest open AI model built using the same research foundation as Gemini 3. However, unlike Gemini, it is designed for developers who want full control over deployment and customization.

This time, Google has gone further than previous releases. Earlier Gemma models were open-weight but still restricted by usage terms. Now, the model is fully open source. That means developers can modify the model and redistribute it commercially with attribution.

As a result, organizations can run AI workloads locally instead of relying on cloud APIs. This shift supports privacy-focused deployments and reduces long-term operational costs.

How does Gemma 4 differ from Gemini?

The difference between Gemma and Gemini is primarily about access and deployment flexibility.

Gemini is Google’s proprietary AI family integrated into products such as Search, Gmail, Docs, and Cloud services. It typically operates through managed environments or subscription layers.

The new model, in contrast, can run directly on user hardware. Developers can integrate it into their own applications without sending prompts or files to external servers.

This local execution model improves privacy and enables offline AI use cases. It also allows teams to control infrastructure decisions independently.

What new capabilities does it introduce?

The new model adds stronger reasoning abilities compared with earlier versions. Google says the model performs better on instruction-following and multi-step logic benchmarks.

It also supports agentic workflows. These workflows allow AI systems to plan tasks and execute steps automatically across structured environments.

In addition, Gemma 4 includes localized coding assistance features. Developers can use the model to generate or interpret code directly within their own systems.

Another upgrade is multimodal processing. The model can now interpret audio inputs and analyze visual content such as charts and diagrams. That expands its usefulness beyond text-only applications.

Together, these improvements position Gemma 4 as Google’s most capable open model so far.

What model sizes are available for developers?

Google released Gemma 4 in four parameter sizes:

2 billion
4 billion
26 billion
31 billion

These variants allow developers to choose between lightweight efficiency and higher performance capability.

Smaller versions can run on edge devices and laptops with compatible GPUs. Larger models are better suited for advanced reasoning workloads or enterprise deployments.

The model also supports long context processing. The largest variants handle context windows of up to 256,000 tokens. Meanwhile, smaller versions support up to 128,000 tokens.

This extended context length enables longer document analysis and deeper conversation memory during inference.

How many languages does Gemma 4 support?

Gemma 4 has been trained on more than 140 languages. This multilingual coverage allows developers to deploy the model across global applications without additional localization layers.

For example, teams building customer support assistants, translation systems, or multilingual analytics tools can integrate Gemma 4 directly into workflows.

This capability also improves accessibility for regional AI deployments outside English-dominant environments.

Why the Apache 2.0 license changes everything

The biggest shift with Gemma 4 is its licensing model.

Google released the model under Apache 2.0, a widely used open-source license that allows redistribution and commercial usage with minimal restrictions.

Developers only need to provide attribution and include the license alongside redistributed versions.

Previously, open-weight AI models often allowed downloads but restricted modifications or commercial reuse. Gemma 4 removes those barriers.

As a result, organizations can build proprietary applications on top of the model without recurring licensing concerns.

Where can developers try Gemma 4 today?

Google has made Gemma 4 available through multiple platforms.

Developers can experiment with it inside Google AI Studio. They can also download model checkpoints from repositories such as Hugging Face, Kaggle, and Ollama.

Because the model supports local execution, teams can deploy it on private infrastructure or edge devices depending on performance requirements.

This flexibility makes experimentation faster and lowers entry barriers for independent developers and startups.

What does this mean for local AI adoption?

The release signals a broader shift toward portable AI models that run outside centralized cloud systems.

Local deployment reduces latency and improves control over sensitive data. It also allows developers to design AI experiences tailored to specific hardware environments.

At the same time, open licensing enables faster innovation across research communities and enterprise ecosystems.

With its reasoning upgrades, multimodal processing, and permissive license, Gemma 4 expands the role of open AI models in production workflows and edge computing scenarios.

In that context, Google’s decision to fully open source Gemma 4 marks an important step toward developer-controlled AI infrastructure.