Meet Gemma-3: Most-powerful AI model to run on single GPU
What's the story
Google has officially unveiled Gemma 3, the latest member of its AI model family for developers. The release comes after the previous versions, Gemma 1 and Gemma 2. It is claimed to be the most powerful AI model that can be run on a single GPU. The company's Gemma family has proven to be extremely popular, racking up over 100 million downloads in just a year and an impressive collection of 60,000 variants dubbed the "Gemmaverse."
Specifications
Why is Gemma 3 special?
Gemma 3 is based on the same research and technology behind Google's Gemini 2.0 models. It is available in four sizes - 1B, 4B, 12B, and a whopping 27B. Google has touted Gemma 3 as the "world's best single-accelerator model," beating competitors like Llama-405B, DeepSeek-V3 and o3-mini in LMArena. The model is optimized to run smoothly on phones, laptops, and workstations.
Features
Advanced reasoning capabilities and language support
Gemma 3 comes with advanced text as well as visual reasoning capabilities, enabling it to analyze pictures, text, and short videos. It supports more than 35 languages out-of-the-box with pre-trained support for over 140 languages. The model also comes with a unique context window of up to 128,000 tokens. These attributes are likely to make Gemma 3 a valuable tool for developers across the globe.
Safety
Enhanced performance and safety features
Gemma 3 brings 'function calling' and 'structured output' capabilities, allowing developers to automate the tasks and build agentic experiences. For safety, Google has created ShieldGemma 2, a robust image safety checker that classifies content into three safety categories: dangerous content, sexually explicit material, or violence.