Inkling: Our open-weights model
Thinking Machines Lab’s first open-weights release is a 975B-parameter multimodal base model built for customization, not frontier performance.
Inkling is Apache-2.0 licensed, with 41B active parameters and training across text, images, audio, and video. Willison notes that the model card and training-data documentation are unusually thin, saying little beyond public/open internet sources, public repositories, and some third-party datasets. Thinking Machines presents it as a strong fine-tuning base for its Tinker platform, with a smaller Inkling-Small still pending release. Willison also tested it on an SVG pelican prompt, then had the model describe its own rendered image. Simon Willison's note
Inkling is Apache-2.0 licensed, with 41B active parameters and training across text, images, audio, and video. Willison notes that the model card and training-data documentation are unusually thin, saying little beyond public/open internet sources, public repositories, and some third-party datasets. Thinking Machines presents it as a strong fine-tuning base for its Tinker platform, with a smaller Inkling-Small still pending release. Willison also tested it on an SVG pelican prompt, then had the model describe its own rendered image. Simon Willison's note
score 9