Thinking Machines Unveils Inkling, an Open-Weight AI Model Built for Customization

Thinking Machines Lab, the AI startup led by former OpenAI CTO Mira Murati, released its first proprietary model Wednesday. Called Inkling, it's an open-weight system - meaning developers and companies can download and modify it directly. That sets it apart from the flagship models of OpenAI, Anthropic, and Google, which are sold as closed, one-size-fits-all products.

Inkling is a mixture-of-experts model with 975 billion total parameters, but it activates only about 41 billion for any given task - a design that keeps large models faster and cheaper to run. Trained on 45 trillion tokens spanning text, images, audio, and video, it can reason across all four formats natively, though its current output is limited to text, including code and structured data. The company says Inkling is designed to give calibrated answers, flagging uncertainty rather than guessing, and lets users dial "thinking effort" up or down for speed. On one benchmark, it uses a third as many tokens as Nvidia's Nemotron 3 Ultra to achieve the same coding performance.

Thinking Machines doesn't claim Inkling is best-in-class. In briefing materials, the company states it is "not the strongest model available today, closed or open." Instead, the pitch is well-rounded performance and customizability - positioning Inkling as a starting point for organizations to fine-tune through Tinker, the company's model-customization platform. That approach is a bet that companies willing to own and adapt their own models can wring more value from them than from centralized, subscription-based AI. The argument is gaining traction: Microsoft CEO Satya Nadella recently warned that enterprises using proprietary AI effectively pay twice - once in subscription costs and again by handing over business knowledge. Hugging Face CEO Clem Delangue similarly predicted most production AI will shift to private or open-source alternatives.

The clearest evidence for Thinking Machines' strategy came from a recent project with Bridgewater Associates, the world's largest hedge fund. Researchers from both companies took an existing open-source model and trained it on Bridgewater's financial expertise. The result scored 84.7% on financial reasoning tests, beating top proprietary models while costing roughly a fourteenth as much to run. Thinking Machines emphasizes it reached this milestone in about nine months - far faster than OpenAI's five years or Anthropic's three. The company now employs roughly 200 people and has struck a strategic partnership with Nvidia for computing capacity. But revenue will likely come from Tinker, not the model itself, since open weights mean no one is obligated to pay Thinking Machines to run them.

Source: TechCrunch