Looking to catch up on all the announcements, deployment strategies, and hands-on guides on “Gemma 4”, here is a reading list to get you up to speed.
Announcement Post
To summarize, most intelligent open model family yet. It can handle handle deep reasoning, multi-step planning, and agentic workflows and fully released under a commercially permissive Apache 2.0 license.
Gemma 4 on Google Cloud
If you are looking to these models into production, Richard Seroter breaks down the myriad of ways you can deploy Gemma 4 on Google Cloud. You can spin it up on Vertex AI for direct infrastructure control, leverage Cloud Run for serverless scaling, use GKE for highly customizable environments, or tap into Google Cloud TPUs for massive scale.
Here is an infographic (courtesy NotebookLM) to help you visualize all the ways you can use these models on Google Cloud.
The Inside Scoop
’s breakdown on LinkedIn, is an excellent summary of the four distinct model variants: the edge-optimized E2B and E4B, the lightning-fast 26B Mixture of Experts (MoE), and the highly capable 31B Dense model. Check out the comments.
Codelabs : Get Hands-On
Read to put Cloud Run & vLLM to test and get Gemma 4 running on Google Cloud? Here is a fantastic codelab that walks you step-by-step through running Gemma 4 inference on Cloud Run with RTX 6000 Pro GPUs using vLLM. It covers the critical networking setups like Direct VPC Egress and configuring vLLM for optimized performance and memory management.
Building Agents
Finally, if you want to see Gemma 4’s reasoning and function-calling in the real world, check out these two deep-dives:
Get Romin Irani’s stories in your inbox
Join Medium for free to get updates from this writer.
shows Java developers how to build an ADK Java agent powered by Gemma 4. He explores three different connection paths: calling the AI Studio API directly, hitting a vLLM-hosted instance via LangChain4j, or running it locally via Ollama.
explores the rise of “Agentic Commerce,” detailing how Gemma 4’s native multimodality, configurable “thinking mode,” and reliable tool-calling make it the most enterprise-ready open model for autonomous retail workflows, with examples spanning from on-device visual search to complex back-office supplier negotiations.
Try Gemma 4 on your mobile
You can try out Gemma 4’s capabilities on the edge too. Access Android’s built-in Gemma 4 model through the new AICore Developer Preview, or leverage Google AI Edge to build agentic, in-app experiences across mobile, desktop, and edge devices.
Check out this announcement blog post that provides instructions on how to do that.