Thinking Machines Lab Releases Open Model 'Inkling'

Thinking Machines Lab, the AI company founded by former OpenAI executive Mira Murati, has released its first model, named Inkling, according to Wired.

Inkling is an open-weight model with 975 billion total parameters. It uses a mixture-of-experts approach that activates about 41 billion parameters for each task, allowing it to operate more efficiently. The model was trained on 45 trillion tokens of text, image, audio, and video content. Although it can reason across these types of data, its current outputs are text-based, covering areas such as code, styles, and structured data.

According to the company, Inkling is built to produce calibrated answers and can signal uncertainty when appropriate. Users can adjust the level of computational effort to balance accuracy and speed. Thinking Machines positions Inkling as a foundation for organizations to customize through its model fine-tuning platform called Tinker.

The release of Inkling follows a year and a half of infrastructure development at Thinking Machines Lab. The company previously previewed its work on interaction models, focusing on AI that can process ongoing conversations instead of the turn-based exchange typical of most chatbots.