Google DeepMind unveiled Gemma 3, a new family of open-source AI models that deliver state-of-the-art performance on consumer hardware. Available in 1B, 4B, 12B, and 27B parameter sizes, these models can run efficiently on a single GPU or TPU, making advanced AI accessible to developers without massive infrastructure.

The 27B variant outperforms larger models like Llama3-405B and DeepSeek-V3 in human preference evaluations, according to LMArena leaderboard tests. Gemma 3 supports over 140 languages, features a 128k-token context window, and includes ShieldGemma 2 for image safety detection across dangerous content, violence, and explicit material.

These models come with official quantized versions for reduced computational requirements and integrate with tools like Hugging Face Transformers, PyTorch, and Google AI Studio. The release builds on Gemma's first birthday milestone of 100 million downloads and 60,000 community-created variants, establishing Google DeepMind as a leader in democratized AI development.