Mira Murati's Thinking Machines releases Inkling open-weight AI model

Thinking Machines released its first AI model Wednesday, marking former OpenAI executive Mira Murati's debut product since founding the startup. The model, called Inkling, represents a strategic bet that enterprises value customization capabilities over pure performance benchmarks.

Why it matters

The release tests whether a $12 billion valuation can be justified by offering enterprises an alternative to renting AI from closed providers like OpenAI and Anthropic. As companies increasingly seek control over their AI infrastructure and intellectual property, open-weight models that can be fine-tuned on proprietary data are gaining traction despite not being the most powerful options available.

Built from the ground up

Thinking Machines constructed Inkling from scratch rather than modifying an existing model, training it on Nvidia's latest AI infrastructure. The company acknowledges Inkling is not the strongest model on the market, instead emphasizing its customizability as a path to better performance at lower costs for specific enterprise use cases.

In its final training phase, the company used data generated by existing open models, including Kimi K2.5 from Chinese lab Moonshot AI. The full weights are now available on Hugging Face, and the model is live for fine-tuning on Tinker, Thinking Machines' customization platform.

The startup is also previewing Inkling-Small, a lighter-weight variant that will be released after testing. Meanwhile, more powerful successors are already in training.

The open-weight advantage

Organizations using open-weight models can fine-tune them on proprietary data and deploy them on infrastructure they control, providing flexibility over hosting and costs that closed models typically don't offer. This dynamic recently gained attention when Palantir CEO Alex Karp criticized frontier AI tools from closed providers as too expensive and lacking clarity on intellectual property protections.

Demand for this approach is growing as companies look for cheaper AI they can customize for specific applications rather than relying on general-purpose models from a handful of labs.

Echoes of OpenAI's pivot

Murati's approach carries historical resonance. She was at OpenAI in 2019 when the lab withheld the full version of GPT-2 over misuse concerns, signaling the company's shift away from fully open releases despite its founding promise of openness.

Thinking Machines has not committed to making all future models open-weight, suggesting a case-by-case strategy that releases models openly when risks are manageable.

Following the capital

Thinking Machines raised a record $2 billion seed round at a $12 billion valuation in 2025 before releasing any product. Nvidia participated as an investor and has since deepened the relationship. The startup also reportedly signed a multibillion-dollar deal with Google Cloud.

Inkling now serves as the first real test of whether that funding can translate into a compelling enterprise alternative, according to Axios, which first reported the details.