Inkling Small

An efficient open-weight model with native image and audio reasoning. Thinking Machines positions it for coding and tool use, while the larger Inkling retains stronger knowledge coverage and factuality.

Inkling Small targets applications that need reasoning and tool use with less compute than the original Inkling. Thinking Machines reports comparable overall performance from a model roughly one quarter its size.

Its training began later, allowing changes to the data mixture and training recipe. The lab reports stronger reasoning and coding results, while the larger Inkling retains an advantage in factuality and knowledge coverage. Small therefore offers a different balance rather than a universal replacement.

Native image and audio processing remain part of the design. The release also improves Python-assisted visual inspection, making charts and documents practical use cases alongside coding. Adjustable thinking effort lets developers trade additional reasoning work against cost.