Europe 2026
Gemma, DeepMind's Family of Open Models — Omar Sanseviero, Google DeepMind
Overview
Gemma is Google DeepMind's family of open models, designed to be downloadable, runnable on personal infrastructure, and fine-tunable for specific use cases. The latest release, Gemma 4, offers models ranging from 2 to 32 billion parameters, with capabilities extending to on-device agentic tasks, multimodal understanding, and multilingual support. These models are developed with a focus on developer-friendly sizes and an Apache 2.0 license for greater flexibility.
Who should watch
- AI engineers and builders interested in open-source large language models.
- Product Managers exploring on-device AI capabilities.
- Developers looking for models that can run offline or on consumer hardware.
- Researchers and practitioners focused on multimodal and multilingual AI applications.
- Anyone interested in the advancements and accessibility of open AI models.
Key takeaways
- Gemma 4 models range from 2B to 32B parameters, with smaller versions capable of running on mobile devices and Raspberry Pi.
- The E2B (effectively 2 billion parameters) architecture utilizes per-layer embeddings for optimized on-device performance, reducing computational load.
- Gemma models support multimodal understanding, including images, videos, and audio, and are heavily multilingual, trained on over 140 languages.
- The models are designed for agentic tasks, coding, and even Android app development, with examples shown running entirely offline on phones.
- Gemma 4 is released under an Apache 2.0 license, offering greater flexibility for developers and users.
- The ecosystem has rapidly adopted Gemma, with over 1,000 community-created models and 10 million downloads of base models within a week of release.
- Google collaborates with open-source tools like llama.cpp, Hugging Face, and Unsloth to ensure Gemma is readily usable within existing developer workflows.
- Gemma has demonstrated potential in scientific research, with one study using Gemma 3 to propose cancer therapy pathways that were validated in a lab.
Notable quotes
Gemma is Google DeepMind's family of open models.
*These are models that you can take, you can download, you can run in your own infrastructure, your own devices.*
*Gemma 4 is the family of most capable of open models that Google has released ever.*
Unofficial community note. Prefer the recording for nuance.