High-Capacity Model: Also with 17B active parameters, but within a larger 400B total, employing 128 experts in its MoE architecture. ​ Versatile Multimodal Understanding: Excels in understanding and generating both text and images across multiple languages, supporting applications like visual question answering and dense summarization. ​ Optimized for Assistant Tasks: Designed for high-quality general assistant and chat use cases, demonstrating strong performance in reasoning tasks.