Google releases new AI model capable of running on a single GPU
Gemma 3 claims to outperform Llama-405B, DeepSeek-V3, and o3-mini in preliminary human preference evaluations on LMArena’s leaderboard. It also provides support for over 35 languages and pre trained support for over 140 languages.
Tech behemoth has introduced Gemma 3, a new series of AI models built to deliver state-of-the-art performance by operating on a single GPU (Graphic Processing Unit).
Built using the same advanced research and technology as the company’s Gemini 2.0 models, Gemma 3 aims to offer developers a portable solution to build AI applications. The newly launched models are also designed to run faster and directly on devices including phones, laptops, and workstations.
The Gemma 3 lineup comes in multiple sizes, varying from 1B, 4B, 12B, and 27B parameters—allowing users to select the model best suited for their hardware and performance requirements.
Gemma 3 claims to outperform Llama-405B, DeepSeek-V3, and o3-mini in preliminary human preference evaluations on LMArena’s leaderboard. The company stated that the model supports over 35 languages and pretrained capabilities spanning more than 140 languages.
With Gemma 3, users can build applications that analyse images, text, and short videos.
The model offers a 128k-token context window, meaning it can analyse much larger blocks of text at once. This allows applications to manage and interpret complex data sets, such as long documents or multi-step processes effectively.
Source: Google
“As the industry develops more powerful models, it will be critical to collectively develop risk-proportionate approaches to safety. We will continue to learn and refine our safety practices for open models over time,” read the company's blog.
The release of Gemma 3 comes at a time when tech rivals are increasingly rolling out sophisticated models, especially after the growing popularity of DeepSeek, an AI model capable of running on lower hardware requirements.
Meta has reportedly begun testing its first in-house chip for training AI (artificial intelligence) models, as it doubles down its efforts to develop its own specialised silicon and lessen its dependence on third-party suppliers like NVIDIA, Reuters reported.
In a similar move, Microsoft is said to be working on its own artificial intelligence models to rival longtime partner OpenAI. The tech giant has reportedly started evaluating AI solutions from xAI, Meta, and DeepSeek as potential alternatives for OpenAI’s technology in Microsoft 365 Copilot, the company’s flagship AI product.
Edited by Kanishk Singh
