MIRA MURATI’S AI COMPANY DROPS ITS FIRST MODEL AND IT IS NOT TRYING TO BEAT GPT — IT IS TRYING TO BE HONEST ABOUT WHAT IT DOES NOT KNOW Mira Murati left OpenAI as its chief technology officer in late 2024, took a year to build, and on Tuesday released the first product from her startup Thinking Machines Lab: an open-weight AI model called Inkling. It is a massive system by any measure, with 975 billion total parameters built as a mixture of experts, meaning it activates only about 41 billion of those parameters at any one time to stay fast and affordable to run. It was trained on 45 trillion tokens spanning text, images, audio, and video, and it can reason natively across all four. What sets it apart from the dominant narrative in AI is what Thinking Machines decided not to claim. The company openly states that Inkling is not the most capable model available today, open or closed. Instead, the model is designed to know what it does not know. It flags uncertainty rather than filling gaps with convincing nonsense, and it lets users dial reasoning effort up or down depending on whether they want speed or depth. Thinking Machines is not trying to monetize Inkling directly. Its revenue comes from Tinker, a fine-tuning tool that hedge funds and large enterprises use to customize AI models for their own data. Inkling is the open bet that the future belongs to specialized, calibrated AI, not the most confident one in the room. Keywords: Thinking Machines, Mira Murati, Inkling model, open weight AI, AI model release