Atlas / Open-weight and challenger labs / Gemma 4
Gemma 4
AI Market Atlas · by Pro Pluribus
Google DeepMind
The strongest open small-model family from a US lab.
What it offers
Main offerings
- Gemma 4
- Sizes from 2B-class edge models to 31B. Multimodal, Apache 2.0, runs on-device.
Recent news
- Google releases EmbeddingGemma 2, a multimodal embedding modelThe Gemma family extends into embeddings under Apache 2.0.AI Weekly ↗
- Google releases Gemma 4 under Apache 2.0, from 2B to 31BBuilt from Gemini 3 research, with strong performance for its size and 256K multimodal contextGoogle Blog ↗
- Gemma 4 comes to Android through the AICore developer previewMakes Gemma the default local agent model on Android devicesAndroid Developers Blog ↗
This map is a by-product of the work. Keeping track of what each of these actually does, and which ones are worth your time, is part of how I advise founder-led companies on AI. If you want that applied to your business, the first conversation is free.
Book a 30-minute conversation →Compiled from public sources by Pro Pluribus; updated 10 Nov 2026. General information, not advice. Company names and logos belong to their owners and are used to identify them, without affiliation or endorsement. Terms