Google has introduced Gemma 3, the latest addition to its open AI model lineup. According to the company’s blog post, Gemma 3 is built on the same research and technology that powers Gemini 2.0 and is available in 1B, 4B, 12B, and 27B sizes. Google claims it is the world’s best single-accelerator model, optimized to run efficiently on a single GPU or TPU. It also outperforms Llama-405B, DeepSeek-V3, and o3-mini in benchmarking tests conducted on LMArena.

What sets Gemma 3 apart is its ability to run efficiently on various devices, from smartphones and laptops to high-performance workstations. Google previously released Gemma 1 in February 2024, followed by Gemma 2 in May. Now, after nearly a year, Gemma 3 arrives with significant improvements in speed, functionality, and safety.

One of its key advancements is its enhanced reasoning capabilities, allowing it to process text, images, and short videos at a scale of over 4 billion parameters. It also features a 128,000-token context window and pre-trained support for over 140 languages, with built-in compatibility for more than 35 languages.

The model introduces AI-powered function calling, enabling developers to automate workflows and create more interactive AI applications. Its officially optimized quantized models offer faster performance and lower computational demands while maintaining high accuracy. Google has also focused on safety, incorporating ShieldGemma 2, a 4B-parameter image safety checker that detects and labels harmful, explicit, or violent content. The company assures that Gemma 3 has undergone rigorous testing to align with its AI safety policies. Developers and users can explore Gemma 3 through Google AI Studio directly from their web browser.

Disclaimer: This image is taken from Business Standard.