Inkling-Small, developed by @thinkymachines, has debuted at approximately rank #88 with 1431 points in Text Arena and rank #21 among open models, making it one of only five US models in the top 25 overall. This early AutoEval score indicates strong initial performance compared to other models, as it employs a Reward Model that simulates human voting based on preference data. With 276 billion parameters, including 12 billion active ones, Inkling-Small is designed to deliver comparable performance to its larger counterpart, Inkling, at a fraction of the size, and full weights are now available for fine-tuning on the Tinker platform.

Tinker: Tinker is a platform designed for fine-tuning AI models and supporting experimentation. It hosts the newly released Inkling-Small for users to customize and work with directly. The service is tied to Thinky Machines’ releases in the open model space.
Inkling: Inkling is a language model created by Thinky Machines that serves as the foundation for smaller variants. Inkling-Small builds directly on this work to extend performance in a more efficient form. The original model supports the ongoing development of open AI resources by the same team.
Inkling-Small: Inkling-Small is a compact language model developed by Thinky Machines as a smaller variant of their Inkling model. It is released with full weights available for community use in the open ecosystem. The model is integrated into Thinky Machines’ Tinker platform for fine-tuning and interaction.
Thinky Machines: Thinky Machines is an AI organization focused on creating and releasing open models to advance the ecosystem. It recently launched Inkling-Small, positioning the release as a contribution to accessible AI development. The organization maintains the Tinker platform and Playground for model engagement.
Tinker Playground: Tinker Playground is an interactive interface provided by Thinky Machines for engaging with models across text, image, and audio modes. It offers immediate access to releases such as Inkling-Small. The playground facilitates user testing and conversation with new open models.

US Open Models: US organizations are delivering competitive open models that rank prominently among global open releases.
Model Accessibility: Platforms supporting fine-tuning and multimodal interaction are expanding options for working with new open AI releases.