Gemma
Verified Oct 2, 2026 ActiveGoogle DeepMind's family of open-weight language and multimodal models, from phone-sized to workstation-sized, free to download and run.
Listed on EffectHub ·
Pricing
- Pricing model
- 100% Free
Model weights are free to download and, for Gemma 4, licensed under Apache 2.0. Running them costs your own hardware or whichever cloud or hosted service you choose; earlier Gemma generations used Google's own Gemma terms of use.
Use tool ↗About Gemma
Gemma is a family of open models from Google DeepMind, built from the same research as Gemini. The current generation, Gemma 4, ships in sizes from small edge models to a 31-billion-parameter model and handles text, images and, in the smaller variants, audio. Specialised variants cover medical text and images, safety classification, encoder-decoder tasks and text diffusion. You download the weights from Kaggle or Hugging Face and run them with tools such as Ollama, LM Studio, PyTorch or JAX. Gemma 4 moved to the Apache 2.0 license; the earlier Gemma 2 listed by many directories is a previous generation.
- API
- No
- Last verified
- Oct 2, 2026
Verified sources
- Gemma 4 sizes, variants and supported platformsdeepmind.google ↗
- Gemma 4 released under the Apache 2.0 licenseopensource.googleblog.com ↗
- First Gemma launch, February 21, 2024blog.google ↗
Is Gemma legit?
We don’t label tools safe or unsafe. Below are the records we checked ourselves, each with the date we read it.
- Last checked
- Oct 2, 2026The site answered
- On the market since
- 2024The vendor's own material, read by us
- Operated by
- Google DeepMind, United StatesOur research, card verified Oct 2, 2026
Pros & cons of Gemma
Gemma is a good starting point when you want an open model you can run locally or fine-tune, with good tooling support and permissive licensing in the latest generation. Pick the size to match your hardware, and treat the small variants as edge models rather than replacements for large hosted ones.
Pros
- Wide range of sizes, from edge models to a 31B model
- Apache 2.0 license for Gemma 4 allows commercial use and modification
- Works with Ollama, LM Studio, Hugging Face, PyTorch, JAX and Android tooling
- Specialised variants for medical, safety and function-calling use cases
Cons
- Weights only: you provide the hardware, serving stack and safety layer
- Older generations such as Gemma 2 use different, more restrictive terms
- Google's page warns the models can produce inaccurate or offensive content
Best for
- Developers who want a self-hosted or on-device model with an Apache 2.0 license
- Researchers and teams fine-tuning small and mid-sized open models
- Mobile and edge projects that need a compact multimodal model
Not for
- People who want a ready-made chat app rather than model weights
- Teams that need the largest frontier-scale reasoning without running their own infrastructure
Watch out: Directories often list Gemma 2, which has been superseded; check the generation and its license before using a specific checkpoint. The terms changed to Apache 2.0 only with Gemma 4.
Top alternatives to Gemma
Gemma head-to-head
Side-by-side fact sheets against the competitors our research or their makers name — pricing, platforms, API and both verdicts on one page.