Lightweight & Efficient: Features 17B active parameters within a 109B total, utilizing 16 experts in a Mixture-of-Experts (MoE) architecture. ​ Extended Context Window: Supports an industry-leading 10 million-token context window, ideal for processing extensive documents like research papers or legal texts. ​ Multimodal Capabilities: Handles text, images, and audio inputs, making it suitable for tasks such as document summarization and visual question answering.