Gemma

Google's free, open AI models small enough to run on a laptop or phone

Changelog

Oct 15
Launch
Gemma launched a new documentation site for developers with a model selector for code snippets, Markdown pages for agents, and in-browser on-device demos.
What this means for you: Gemma's new documentation site is a place for developers to learn and try the models. It has a model selector that customizes code snippets, and any page can be turned into Markdown, a plain text format that AI agents read easily. It also has demos that run on your own device inside the browser.
Why it matters: It follows a run of Gemma updates for developers, including Gemma 4 running in Colab on Oct 9 and EmbeddingGemma 2 on Oct 6.
How to use it: Visit gemma.ai. The on-device demos run directly in your browser.
@googlegemma on X
Oct 9
Availability
Gemma 4 models can now be run and fine-tuned in Colab on A100 GPUs with 80GB memory through Google AI Pro plans, and on H100s with Ultra.
@googlegemma on X
Oct 6
New model
Gemma released EmbeddingGemma 2, a 740M-parameter multimodal embedding model for text, code, images, video and audio, with an 8K context window and Apache 2.0 license.
@googlegemma on X

Sep 23
New feature
Gemma 4 and LiteRT now run locally in the Antigravity SDK, letting developers build agents offline, with support for OpenAI-compatible endpoints like Ollama and vLLM.
@googlegemma on X
Sep 16
Availability
Gemma 4 12B is now fully supported in the Google AI Edge Gallery app on Mac, with vision and audio input, and runs on 16GB Macs.
@googlegemma on X
Sep 2
Availability
Gemma 4 is now the recommended local model provider in Android Studio, letting app developers use an offline coding agent with no token quotas.
@googlegemma on X