Richard Lawler is a senior journalist who follows news in tech, culture, politics, and entertainment. He joined The Verge 2021, after several years of covering news for Engadget.
After releasing two “open Gemma AI” models, built on the same technology as its Gemini AI model, Google is updating this family with Gemma 3. According to the blog, these models were designed for developers to create AI applications that can run anywhere, from a phone or workstation, with support for 35 languages and the ability to analyze images, text, and short videos.
According to the company, it is the “world’s most advanced single-accelerator models”outperforming competitors such as Facebook’s Llama and DeepSeek. It also has optimized capabilities for running Nvidia GPUs and dedicated AI Hardware. Gemma 3ās vision encoder has also been upgraded to support high-resolution and non-square images. The ShieldGemma 2 Image Safety Classifier can be used to filter both input and output of images for content classified as violent, dangerous or sexually explicit. You can read the 26-page reportto learn more about these claims.
It was not clear last year how much interest would be shown in a model such as Gemma. However, the popularity and demand for AI tech that requires less hardware is evident. Google claims that Gemma 3 has advanced capabilities but also states that “Gemma 3ās enhanced STEM performance has prompted specific evaluations focused upon its potential misuse in creating harmful chemicals; their results indicate that there is a low-risk level.”
What constitutes an open or “open source AI model” remains a subject of debate. With Google’s Gemma this has been focused on the license that restricts how people can use it, and that has not changed with the new release. Google continues to promote Gemma by offering Google Cloud credits. The Gemma 3 Academic Program will allow academic researchers the opportunity to apply for $10,000 in credits to accelerate their own research.
