Google releases Gemma 3: Multimodal, Multilingual LLM with 128K Context

AI Impact Summary

Google has released Gemma 3, a new open-weight multimodal LLM with significant advancements including a 128K token context window, support for image and text inputs, and multilingual capabilities across 140+ languages. The model’s technical enhancements, such as optimized KV cache management and adjusted positional embeddings, enable longer context lengths and improved performance compared to previous Gemma models like Gemma 2. This release represents a substantial upgrade in capabilities and is positioned to compete with leading closed models.