Inkling: Our open-weights model

This article covers the release of Inkling, an open-weights multimodal AI model by Mira Murati's Thinking Machines Lab. Inkling is a Mixture-of-Experts transformer with 975B total parameters (41B active), trained on 45 trillion tokens of text, images, audio, and video. It is Apache-2.0 licensed and intended as a strong base model for fine-tuning via the Tinker platform. The article also mentions Inkling-Small (276B parameters) still in testing, compares Inkling to other open-weight models like NVIDIA Nemotron and Gemma 4, and includes examples of using the model via API to generate SVG images and describe them. The model card is noted for being brief, with minimal training data documentation.