Inkling

Inkling is Thinking Machines’ open-weights multimodal model, designed for broad applicability and customization. Key features include: • 975B MoE with 41B active parameters • 1M token context window • Native reasoning across text, images, and audio • Controllable thinking effort for balancing cost and performance • Apache 2.0 license for full weight access This model provides a robust foundation for developers and organizations looking to integrate advanced understanding capabilities into their systems. It handles complex input combinations from various modalities, allowing for rich data processing and insight generation. The Mixture-of-Experts architecture ensures efficient resource utilization, enabling scalability across diverse applications. Inkling prioritizes flexibility and ease of adaptation. Its open-weight nature, combined with the option to fine-tune on the Tinker platform, empowers users to tailor the model specifically to their unique requirements and datasets. This makes it an ideal choice for building specialized applications without proprietary lock-in. Built for developers, researchers, and enterprises, Inkling supports use cases ranging from advanced content understanding and detailed data analysis to interactive system development. Its emphasis on accessibility and customization makes it a valuable asset for anyone seeking to extend human judgment and create bespoke solutions.