In the latest from the world of AI, Google has just unveiled Gemma 3, a collection of lightweight, state-of-the-art open models built on the same research and technology that powers the Gemini 2.0 models. According to Google, Gemma 3 models are their most advanced and portable open models, capable of running on anything from phones and laptops to workstations.

This launch comes at a time when open AI models are being praised as fantastic and world-class by industry experts, with DeepSeek’s R1 being the most notable example. The Chinese startup's model surged in popularity thanks to its open-source approach, lower development costs, and efficiency in computational resources. So, naturally, we had to compare the newest addition to the open AI space with one of the strongest contenders. And, spoiler alert, Gemma 3 is close.

Gemma 3 ranks as the second-best open model AI

Google’s Gemma 3-27B has 27 billion parameters, which might sound like a big number, but is a lot less compared to other high-ranking models like Meta’s Llama 3.1, which boasts 405 billion parameters. Despite having fewer parameters, it still ranks in the top 10 overall with an Elo score of around 1339 on Chatbot Arena. Additionally, Gemma 3 comes in a range of sizes (1B, 4B, 12B, and 27B), allowing you to choose the best model for your specific hardware and performance needs.

All of this makes Gemma 3 the second-best open model AI, just behind DeepSeek’s R1, which pulls ahead with 25 more points at an Elo score of 1,363. However, while the Chinese startup leads the competition, its model comes with a significantly larger parameter size of 671 billion. On top of that, Gemma 3 achieves its performance using just a single Nvidia GPU, whereas DeepSeek’s R1 is estimated to require up to 32 GPUs.

Built-in image safety checker

There’s a lot more that Gemma 3 comes with, like a 128K expanded context window, faster on-device inference, support for over 140 languages, and even multimodal understanding, so we’d highly suggest checking out Google’s blog to get a complete idea of its capabilities. However, one other highlight is ShieldGemma 2, which is a powerful built-in image safety checker that outputs safety labels across three categories: dangerous content, sexually explicit material, and violence.

That said, Gemma 3 marks a significant leap forward compared to where the company stood with Gemma 2-27B thanks to improved efficiency, a higher ranking on Chatbot Arena, and expanded capabilities. This shows that, just like other major AI companies that have been in high gear since DeepSeek's launch, Google is also making sure it doesn’t fall behind.

About the Author

Hassam Nasir

Hassam boasts over seven years of professional experience as a dedicated PC hardware reviewer and writer.