Google Introduces Gemma 3: A Lightweight Open AI Model for Efficient Processing
| Aspect | Details |
|---|---|
| Why in News? | Google introduced Gemma 3, a lightweight open AI model for efficient performance on smartphones, laptops, and computing platforms. |
| Key Features | - Multi-modal processing (text + visual input, text-only output). - Scalable model variants (1B, 4B, 12B, 27B parameters). - 128k-token context window for better comprehension. - Supports 140+ languages. |
| Training & Tokens | - 1B model: 2 trillion tokens. - 4B model: 4 trillion tokens. - 12B model: 12 trillion tokens. - 27B model: 14 trillion tokens. |
| Performance | Outperforms Meta's Llama-405B, OpenAI's o3-mini, and DeepSeek-V3 on LMArena benchmarks. Achieves higher human preference scores and supports 35+ languages. |
| Use Cases | - Automated translation and multilingual chatbots. - AI-powered automation and intelligent virtual assistants. - Content moderation, video summarization, and data analytics. |
| Availability | Available on Kaggle, Hugging Face, and Google Studio. |
| Deployment Options | - Vertex AI, Cloud Run, Google GenAI API, and local environments. |