Google goes lightweight with new AI models
Add Axios as your preferred source to
see more of our stories on Google.

Illustration: Sarah Grillo/Axios
Google released Gemma, a range of "lightweight" open AI models designed for text generation and other language tasks.
Why it matters: Google is betting on the substantial market of developers who don't need or can't afford to use the biggest AI models like Gemini.
Details: Google is releasing models in two sizes, and both can run on a laptop.
- The models come with a "Responsible Generative AI Toolkit," which Google says will help developers build their own safety filters for Gemma models, and a "debugging tool" to help developers investigate Gemma's behavior and address potential issues.
- Gemma is optimized for Nvidia GPUs, offering the ability to fine tune models locally.
- Access to Gemma is free via Google's Kaggle platform for data scientists in an effort to encourage transparency about how the models are used and to offer "large scale community validation" of its safety efforts.
The big picture: Instead of chasing an excitement factor or a consumer market, Google is seeding an enterprise market — one that may end up paying big dollars to use Google Cloud, as developers invent new consumer applications to run on Gemma.
What they're saying: "Things that previously would have been the remit of extremely large models are now possible with state-of-the-art smaller models and this unlocks completely new ways of developing AI applications," said Tris Warkentin, a director at Google DeepMind.
Yes, but: Microsoft has also invested in the market for smaller models, via its Phi range.
- While Google emphasized Gemma's "strict terms" of responsible use, it's placing no limits on which organizations may use the model, creating the risk that malicious actors repurpose the model for unintended uses.
Go deeper: How competition between big and small AI will shape the tech's future
