Thinking Machines launches Inkling as AI rivalry deepens

Thinking Machines launched Inkling, its first AI model, putting Mira Murati’s startup at the center of a sharper open-source AI policy fight.

The release gives the company its first public product after a $2 billion funding round last year, when it had no product or revenue. Inkling is described as less capable than the leading systems from both U.S. and Chinese developers, but its design choices make the debut more important than raw performance alone.

Inkling starts below leaders

Murati, a former OpenAI executive, founded Thinking Machines Lab after leaving one of the companies most associated with closed, frontier AI systems. Her new firm’s first model moves in a different direction: Inkling is open-source, a less common approach among high-profile U.S. AI model releases.

The model also draws attention because its architecture and training data are described as taking cues from Chinese competitors. That combination puts Thinking Machines into a policy argument that is no longer only about who has the strongest model, but about how model knowledge travels across borders.

Chinese methods enter U.S. debate

Open-source AI allows outside developers to inspect, adapt and build on released code or model components, depending on the license and technical package. Supporters view that as a way to widen access and speed research; critics warn it can also spread capabilities faster than governments can regulate them.

Washington’s concern has sharpened around powerful Chinese open-source models, which can be used widely at lower cost than closed systems. The source material says U.S. proposals under discussion do not address what Reed Albergotti called the “true problem”: Chinese firms allegedly using U.S.-made frontier models to train their own systems through illicit distillation attacks.

Distillation generally refers to training a smaller or newer model to mimic outputs from a stronger one. If done without permission against a protected frontier system, it can transfer some capabilities while avoiding the full cost of original training, undercutting the defensive moat around expensive U.S. models.

A $2 billion credibility test

For Thinking Machines, Inkling changes the company’s market position from promise to proof. The $2 billion round gave Murati’s team unusual expectations before revenue or a product launch, and Inkling now becomes the first tangible basis for judging the lab’s technical path.

The early signal is mixed. An open-source release can attract developers, researchers and enterprise testers, but a model that trails top U.S. and Chinese systems must compete on usability, transparency, cost and community adoption rather than headline capability.

Three paths for Inkling

If Inkling gains developer traction despite weaker performance, Thinking Machines could build a distribution advantage before it has a frontier-class model. The macro effect would be a wider diffusion of AI tools; the company would gain credibility with users; the industry would face more pressure to defend closed-model pricing.

If Washington focuses on restricting Chinese open-source systems or distillation abuses, the near-term effect would be greater compliance friction across the AI supply chain. Thinking Machines would need to show that Inkling’s training inputs and model design can withstand scrutiny, while the wider sector would likely spend more on provenance, audits and usage controls.

If Inkling fails to close the capability gap, the $2 billion valuation story becomes harder to support. That would have limited direct macro impact, but it could cool investor appetite for pre-revenue AI labs and push the industry toward clearer proof points: paying customers, defensible data access and models that justify their training costs.