Gemma 3 delivers state-of-the-art performance for its size, outperforming Llama-405B, DeepSeek-V3 and o3-mini in preliminary human preference evaluations on LMArena’s leaderboard
By Ada Jain
Google announced on Wednesday the launch of Gemma 3, a collection of lightweight, state-of-the-art open-sourced AI models built from the same research and technology that powered their Gemini 2.0 models. These are believed to be their most advanced, portable and responsibly developed open models yet for more reasons than one.
“We listened to community feedback and added the most requested features, such as longer context, multimodality, and more!” Google said in a statement, pointing to the newest iteration to the Gemma models.
Gemma 3 offers improved math, reasoning, and chat capabilities with a score of 1338 in LMArena—a platform used for evaluating AI models through human preferences—making it a top open compact model. A key difference that sets it apart from Gemini is that it allows for higher adaptability, while the latter is not designed for developers to fine-tune the models for specific tasks or datasets.
READ: Google’s AI chatbot Gemini faces accuracy concerns (December 20, 2024)
Gemma 3 delivers state-of-the-art performance for its size, outperforming Llama-405B, DeepSeek-V3 and o3-mini in preliminary human preference evaluations on LMArena’s leaderboard. This helps to create engaging user experiences that can fit on a single GPU or TPU host, ideal for startups and individual developers working on budget constraints as well as keeping ethical considerations of AI at the forefront through Google’s Responsible Generative AI toolkit.
Alongside Gemma 3, they also launched ShieldGemma 2, a powerful 4-bit image safety checker built on the Gemma 3’s foundation. ShieldGemma 2 provides a ready-made solution for image safety, outputting safety labels across three safety categories: dangerous content, sexually explicit and violence. Developers can further customize ShieldGemma for their safety needs and users.
However, this new launch is just one slice of the pie to democratize AI technology for developers as Google also launched the Gemma 3 Academic Program. Academic researchers can now apply for Google Cloud credits (worth $10,000 per award) to accelerate their Gemma 3-based research. The application form opens today, and will remain open for four weeks on their website.
READ: Apple eager to defend Google during antitrust trial (December 26, 2024)
Some of the success stories in the “Gemmaverse”—a community of tools designed with Gemma—include AI Singapore’s SEA-LION v3 which breaks down language barriers and fosters communication across Southeast Asia; INSAIT’s BgGPT, a pioneering Bulgarian-first large language model that demonstrates the power of Gemma to support diverse languages; and Nexa AI’s OmniAudio showcases the potential of on-device AI, bringing advanced audio processing capabilities to everyday devices.