Model
Gemma 4.
Power
Varies
API cost
In plain English
Google’s current family of open-weight models. Gemma can run on company infrastructure, personal computers, and some mobile devices, rather than only through Google’s online services.
Good for
Private or offline AI, custom applications, smaller devices, and organisations that want more control over where a model runs.
Things to be aware of
Gemma 4 comes in several sizes, and they do not perform equally. The smaller versions need less computing power but give up capability. If you run it yourself, updates, security, monitoring, and support become your responsibility.
What Google DeepMind says
“Our most intelligent open models, built from Gemini 3 research and technology to maximize intelligence-per-parameter.”
Seen here: Google DeepMind on Gemma 4 · Google DeepMind
What everyone else says
Independent reviewers like how much Gemma can do on comparatively modest hardware. They also warn that its smaller versions are much weaker and that the largest hosted models still perform better on the hardest work. Independent Gemma 4 review · gemmai4.com
- Context window
- 128.0K to 256K tokens
- The context window depends on the version. The two smallest models support 128K; the larger models support 256K. ai.google.dev
- Knows up to
- January 2025
- Run it yourself
- Yes, downloadable weights