A real Gemma 4 playground with optional reasoning traces. Trusted run‑locally and download guides (Ollama, LM Studio, llama.cpp, MLX, vLLM, Unsloth). Not official. Privacy‑aware and transparent.
This is an independent community site. Studio currently uses the logged-in chat experience.
Start here
Release: March 31, 2026
Announcement: April 2, 2026
31B, 26B, and E4B
Download, requirements, and local setup
Release date
March 31, 2026
Announcement post
April 2, 2026
If you are choosing a model
31B is the strongest, 26B is the balanced choice, and E4B is the easiest place to start for lighter local use.
If you want to try it now
Open Studio, or jump into the Ollama guide for the fastest local setup.
Quick links
Most visitors are looking for a model size, download steps, hardware requirements, or a local runtime. Start there.
High-capability tier for demanding reasoning and tool use.
Balanced option for capable local or hosted use.
Smallest practical starting point for edge and laptop testing.
Memory, hardware, and deployment expectations before you commit.
Official release channels, weights, and local ecosystem entry points.
The fastest path to a practical local Gemma 4 session.
Gemma 4 vs Qwen and vs Gemini—decide by task, cost, and speed.
Choose a size
If you are unsure, start smaller and move up. The best model is the one that fits your hardware, speed needs, and everyday tasks.
31B
Choose 31B when output quality matters more than speed and you can afford bigger hardware or hosted inference.
Hardware
Large local GPUs or hosted inference.
Best for
Hard reasoning tasks, longer structured outputs, best-quality comparisons.
26B
Start here if you want strong results without jumping straight to the largest model.
Hardware
Good local rigs or managed inference.
Best for
Coding help, research, daily assistant work.
E4B
Choose E4B if you want the easiest local starting point.
Hardware
Laptops, edge devices, smaller test setups.
Best for
Prompt testing, prototypes, lightweight workflows.
Run locally
Use Ollama for fast CLI setup, LM Studio for a desktop UI, llama.cpp for GGUF workflows, MLX for Apple silicon, vLLM for serving, and Unsloth for tuning.
Compare models
These comparison pages help you decide quickly between Gemma 4, Qwen, and Gemini.
Try it online
Studio is the current logged-in chat experience. Use it to compare models, draft prompts, and turn notes into something you can actually run.
Studio prompts
Compare Gemma 4 31B and 26B for coding plus daily local use.
Write an Ollama-first setup checklist for Gemma 4 on a Mac mini.
Turn these notes into a cinematic video prompt and a short voiceover outline.
Create next
Once you have a clear prompt, brief, or workflow, move into video or music tools. The homepage stays focused on helping you choose and set up Gemma 4 first.
Use Studio or the guides to decide what to run and how to run it.
Turn the answer into a reusable prompt stack, script, checklist, or creative brief.
Move that structured output into video or music generation when the direction is clear.
Use Studio outputs as a shot list, prompt stack, or short script before opening video tools.
Open Video CreatorConvert mood, pacing, and scene notes into something ready for music generation.
Open Music CreatorAbout this site
gemma-4.org is an independent community site. We link to official Gemma materials, publish practical guides, and offer a community Studio on top.
Independent community site. Not an official Google product.
Guides are based on official release notes, model cards, and integration docs.
Studio uses account-based chat. Do not expect an anonymous playground.