Inkling Small (High)
Inkling Small is a sparse Mixture-of-Experts (MoE) multimodal model developed by Thinking Machines Lab as a lighter-weight sibling to their flagship Inkling model. Built with 276 billion total parameters and activating approximately 12 billion parameters per token, it delivers high performance across complex reasoning, coding, and multi-turn conversational tasks while maintaining low latency and efficient compute costs.
Architecture and Capabilities
The model utilizes a 42-layer decoder-only transformer architecture incorporating a sparse MoE feed-forward backbone with 256 routed experts and 2 shared experts per layer, where 6 experts are active simultaneously for each token. Inkling Small natively processes text, image, and audio inputs and projects them into a shared hidden space, supporting a massive context window of up to 1 million tokens. Furthermore, it incorporates a controllable reasoning effort mechanism, enabling users or applications to dynamically program the thinking budget prior to response generation.
Create with Crafiq
Generate images, 3D models, video and audio in one studio.
Explore the studioHow Inkling Small (High) ranks
Inkling Small (High) is highlighted in the table below. Switch the metric to see how the ordering changes.