Google on Wednesday launched its latest Gemma 3 series of artificial intelligence (AI) models. This series is an advanced version of the Gemma 2 series introduced in August 2024. These open-source models come with text and visual reasoning capabilities. The Mountain View-based technology giant said that these models support more than 35 languages ​​and can be fine-tuned to provide additional support for 140 languages. The special thing is that these models can work smoothly on a single GPU or Google’s Tensor Processing Unit (TPU).
Google launches Gemma 3 series AI models
Google gave information about these new Small Language Models (SLM) in a blog post. The company said that these models are based on the same technology used in their flagship Gemini 2.0 model. The Gemma series is known for its open-sourceness and on-device performance. According to Google, the Gemma models have been downloaded over 100 million times so far and over 60,000 variants have been created using them. Read Also: Spectacular new trailer for Death Stranding 2 reveals release date, gameplay, and Metal Gear influences
Google claims that the Gemma models have outperformed Meta’s Llama-405B, DeepSeek-V3 and OpenAI’s o3-mini AI models on the LMArena leaderboard. These models are available in four different sizes – 1B, 4B, 12B and 27B parameters. These models can be run on a single GPU or TPU. The Gemma 3 series has advanced text and visual reasoning capabilities, allowing them to analyze images, text and short videos.
Key features of the Gemma 3 series
- Multi-language support: Support for over 35 languages ​​and fine-tuning options for 140 languages.
- Advanced capabilities: Text and visual reasoning, allowing analysis of images, text and short videos.
- High performance: Available in 1B, 4B, 12B and 27B parameters, which can be executed on a single GPU or TPU.
- Context window: Capacity of up to 1,28,000 tokens.
- Function calling support: Allows developers to add agentic capabilities to apps and software.
Security and optimization
Google said that these AI models have undergone a rigorous risk assessment process. The company used fine-tuning and benchmark evaluation to enforce internal security policies. These models show low risk levels, making them safe to use.
Additionally, Google has also launched ShieldGemma 2. It is a 4B parameter image safety checker, which ensures that AI models do not generate any dangerous, sexually explicit or violent content. Developers also have the option to customize ShieldGemma according to their needs.
The Gemma 3 series AI models can be downloaded from Google’s Hugging Face listing or Kaggle.
