Skip to content

Gemma

Gemma is Google DeepMind's family of open models, positioned for developers who want to build responsible AI applications at scale. It sits alongside the Gemini API and Google AI Studio in Google's developer stack, and the official pages describe it as a unified, encoder-free multimodal model.

260

Overview

Gemma is Google DeepMind's open model family, aimed at developers who want to build responsible AI applications at scale rather than only call a hosted endpoint. It appears in the DeepMind models navigation under open models, separate from the Gemini line, and it has its own page at ai.google.dev/gemma. The official description calls it a unified, encoder-free multimodal model, which is the most concrete technical claim the pages make about it.

Where Gemma Sits in Google's Model Lineup

Google's model navigation splits into several groups. Gemini, Gemini Omni, Nano Banana, and Gemini Audio cover agent building, general creation, image editing, and audio. Veo, Lyria, Genie 3, and Gemini Robotics handle video, music, interactive worlds, and robotics. Gemma is the only entry listed under open models, which is the clearest signal of how Google positions it: as the open counterpart to the hosted Gemini family. That distinction matters for teams that need to run or adapt a model themselves instead of relying entirely on a managed API.

The Encoder-Free Multimodal Design

The phrase "unified, encoder-free multimodal model" is worth reading carefully. Many multimodal systems pair a separate vision or audio encoder with a language model. An encoder-free design folds that processing into a single model, which can simplify the pipeline and reduce the number of components a team has to manage. The official pages do not go further into architecture, benchmarks, or modality-by-modality performance, so anyone evaluating Gemma for a specific workload should treat this as a starting point and check the model documentation for details.

Working Alongside the Gemini API and AI Studio

Gemma does not exist in isolation. Google's developer stack includes the Gemini API, Google AI Studio, and a long list of capabilities such as text generation, document processing, structured outputs, function calling, long context, embeddings, and live audio. AI Studio usage is described as free of charge in all available regions, which makes it a practical place to experiment before committing to a production path. For teams already building on Google's tooling, Gemma is an additional option rather than a separate ecosystem.

What the Pricing Pages Actually Cover

The pricing documentation reviewed here covers the Gemini Developer API, not Gemma. It lists model tiers such as Gemini 3.8 Flash, Gemini 3.6 Flash, Gemini 3.8 Live, Gemini 3.5 Live Translate, and Gemini 3.5 Transcribe Live, along with notes on agentic video understanding, document token billing, and grounding charges. Gemma pricing is not stated on these pages. Any cost planning for Gemma should start from the official Gemma documentation and licensing terms rather than from the Gemini API price list.

A Reasonable Fit for Open-Model Projects

Gemma makes the most sense for teams that want an open foundation from a major lab and are comfortable working within Google's developer ecosystem. The combination of an open model family, a documented page on ai.google.dev, and proximity to the Gemini API and AI Studio lowers the friction of getting started. It is less suited to buyers looking for a fully managed, turnkey service with published per-token rates, since that is what the Gemini line provides.

What to Verify Before Committing

The official pages are thin on specifics. They do not list parameter counts, context windows, supported languages, benchmark results, or licensing terms for Gemma. Before adopting it, check the model card, the license, and the current documentation for the version you intend to use. Treat the encoder-free multimodal description as a design statement, not a performance guarantee, and confirm that the model's actual capabilities match your workload.

What is Gemma? Google's open model family in plain terms

Gemma is a family of open-weight AI models developed by Google DeepMind. Unlike Gemini, Google's flagship models, which you can only use through Google's hosted services, Gemma's model weights are published for download — so developers can run, fine-tune and ship the models on their own hardware, from a phone to a single-GPU workstation to a cloud server. The two families share DNA: Gemma is built from the same research and technology as Gemini, packaged into smaller models that leave Google's data centre.

Gemma 3 (March 2025) is the generation most projects still run. It comes in 1B, 4B, 12B and 27B sizes, plus a tiny 270M variant and the on-device Gemma 3n line. The 4B and larger models are multimodal — they take text and images in and produce text — with a 128K-token context window (the 1B is text-only with 32K), and the family supports more than 140 languages. Gemma 3 is distributed under Google's own Gemma Terms of Use, which permit commercial use but are not a standard open-source licence.

Gemma 4 (2026) is the current generation: effective 2B and 4B models for phones and edge devices, a 26B mixture-of-experts model and a 31B dense model, with a 12B unified variant added later. Built from Gemini 3 research, it adds audio input alongside text and images — and it changed the licensing story, moving the whole family to the standard Apache 2.0 open-source licence.

That licensing detail is the one nuance worth remembering. "Open" for Gemma has always meant open weights: free to download, free to build on, under terms that vary by generation. Weights are available from Hugging Face and Kaggle, and the models also run in Google AI Studio and on Vertex AI if you would rather not host anything yourself.

Frequently asked questions

What is Gemma? Gemma is Google DeepMind's family of open-weight AI models. Google publishes the model weights for download, so anyone can run and fine-tune them on their own hardware — from phones and laptops to cloud GPUs — instead of calling a hosted API.

What is Google Gemma used for? Developers use Gemma to build chatbots, summarisation and classification tools, on-device assistants and research projects where they need a capable model they fully control. The smaller sizes suit phones and edge devices; the 12B and 27B sizes target workstations and servers.

Is Gemma free and open source? The weights are free to download and use, including commercially. The licence depends on the generation: Gemma 3 and earlier use Google's Gemma Terms of Use (open weights, but not an OSI open-source licence), while Gemma 4 is released under the standard Apache 2.0 open-source licence.

What is the difference between Gemma and Gemini? Gemini is Google's proprietary frontier model family, available only through Google's APIs and products. Gemma is the open counterpart — smaller models built from the same research and technology, with weights you can download, run locally and fine-tune yourself.

Core features

  • Open model family from Google DeepMind, described as built for responsible AI applications at scale
  • Positioned as a unified, encoder-free multimodal model
  • Listed under open models on the DeepMind models navigation
  • Documented at ai.google.dev/gemma with a dedicated model page
  • Connected to the broader Gemini API and Google AI Studio developer tooling
  • Sits alongside Gemini, Gemini Omni, Nano Banana, Gemini Audio, Veo, Lyria, Genie, and Gemini Robotics in the model lineup

Use cases

  1. 1Building AI applications on an open model foundation
  2. 2Multimodal work where an encoder-free architecture is relevant
  3. 3Teams standardizing on Google's developer stack across Gemini API and AI Studio
  4. 4Projects that need to weigh open models against the hosted Gemini models
  5. 5Prototyping in Google AI Studio before moving to production

Strengths and things to check

Strengths

  • Backed by Google DeepMind and documented on official Google developer pages
  • Open model positioning gives teams more control than a closed hosted model alone
  • Multimodal, encoder-free design is a distinct architectural choice
  • Fits into an existing ecosystem that includes the Gemini API and Google AI Studio
  • Clear official entry point at ai.google.dev/gemma

Things to check

  • The official pages provide limited technical detail on Gemma itself, so capability claims should be verified against model cards and documentation
  • Pricing for Gemma is not stated on the pages reviewed; the pricing page covers Gemini API models instead
  • Model names and versions on the pricing page may change, so confirm current availability before committing
  • Open model terms and licensing should be checked directly rather than assumed

Pricing

Subscription pricing is not listed.

Basic information

Added
Oct 10, 2026
Platforms
Web
Chinese & English
Not marked
API
Marked as supported
Self-hosting
Marked as supported
Best for
Developers and machine learning teams who want open models they can build on, including those already working within the Gemini API and Google AI Studio ecosystem.
Information updated
Nov 10, 2026
Website status
Checking
Corrections · Privacy · Data requests?
Contact us

Features, pricing and availability may change. Always verify details on the official website. Maintained by steven · Last updated Nov 10, 2026 →

More on this tool

Related tools

All categories