RT by @huggingface: Today, we are releasing Inkling-Small. Inkling-Small achieves comparable…

TL;DR - Thinking Machines released Inkling-Small, an open-weights multimodal MoE model that reportedly matches its larger Inkling sibling at roughly a quarter of the size. It matters as another frontier-lab open-weights drop with fine-tuning and multimodal chat available out of the box.

• 276B total parameters with only 12B active per token (Mixture-of-Experts), targeting large-model quality at much lower inference cost.
• Claimed performance comparable to the full Inkling model despite ~4x smaller footprint; no specific benchmark numbers are given in the post.
• Full weights are being released openly, plus support for fine-tuning via Tinker.
• Multimodal by design — text, image, and audio interaction in Tinker Playground — with "controllable reasoning effort" as a stated feature.