Inkling-Small Weights Released: 276B Total Params with 12B Active

clarejtbirch · x · 2026-07-31

Thinky Machines has released the full weights for Inkling-Small. Utilizing a Mixture-of-Experts (MoE) architecture, the model has 276B total parameters with only 12B active during inference. The developers claim it achieves performance comparable to the larger Inkling model.

Users can now fine-tune the model on the Tinker platform or interact with it via text, image, and audio in the Tinker Playground.

Related event: Thinking Machines Unveils Inkling-Small Multimodal Model(36 posts)→

Original post →

More from Models

Models channel →