Inkling is an open-weights foundation model released by Thinking Machines Lab, positioned as a flexible base that people can adapt to their own needs rather than as the strongest model on the market. It is a Mixture-of-Experts transformer with 975B total parameters and 41B active per inference, designed to balance capability with efficiency. Training was deliberately broad and balanced: the model was pretrained on 45 trillion tokens spanning text, images, audio, and video, which gives it native multimodal reasoning across those domains and supports a context window of up to one million tokens for long, cross-format workflows.

Inkling represents the first entry in a planned family of models at different sizes, with a lighter Inkling-Small preview shared alongside it for lower-latency and lower-cost use cases. The full model emphasizes controllable thinking effort, multimodal understanding, and open-weight availability as practical strengths for teams that want to fine-tune, deploy privately, or experiment without depending on a closed API. Its fit lies in being a broad, customizable starting point: well suited to projects that need an adaptable open base with strong multimodal grounding and the ability to scale down to a smaller sibling when efficiency matters more than peak capability.