Thinking Machines Lab released Inkling on July 15, its first foundational AI model. Founded by former OpenAI CTO Mira Murati, the company raised a record $2 billion seed round at a $12 billion valuation in 2025 before shipping any product. Inkling is the first test of whether that money translates into something the market wants.
What Inkling Actually Is
Inkling is an open-weights model with 975 billion parameters. That puts it well below the estimated scale of frontier models from OpenAI (GPT 5.6) and Anthropic (Claude Fable 5), but Thinking Machines is not trying to win on raw benchmarks. The pitch is customization: enterprises can take Inkling, fine-tune it on their own data, and run it on their own infrastructure.
The model was trained from scratch on Nvidia’s latest AI hardware, and the training process used data generated by other open models, including Moonshot AI’s Kimi K2.5. That last detail is notable: a US startup using Chinese open-source model outputs to train its own model highlights how interconnected the global AI ecosystem has become.
Open Weights, Not Open Source
Inkling ships with open weights, meaning developers can download the model parameters, modify them, and deploy the model without paying licensing fees. This is different from truly open-source models where the training code and data are also released. Thinking Machines has kept those proprietary.
The model handles text, images, and audio natively, and supports multimodal reasoning. It is designed for enterprise use cases where data privacy matters: companies can run Inkling on-premises instead of sending sensitive data to a cloud API.
The Nvidia Relationship
Nvidia was an early investor in Thinking Machines and has deepened the partnership since. The model was trained on Nvidia’s latest AI infrastructure, which gave Thinking Machines early access to hardware that most companies cannot get. Nvidia also benefits from promoting models that run well on its chips, creating a virtuous cycle for both companies.
Thinking Machines reportedly signed a multibillion-dollar deal with Google Cloud as well, giving it cloud distribution alongside its on-premises focus.
What Comes Next
Murati has indicated that Inkling is just the beginning. The company is already training more powerful successor models that may or may not be released openly. Her previous experience at OpenAI, where she oversaw the release of GPT-4 and led the DALL-E project, gives Thinking Machines credibility that few startups can match.
The broader strategy mirrors what Mistral and Meta have done with their open models: release capable but not frontier models openly to build a developer ecosystem, then monetize through enterprise services and larger closed models.
The open-weight AI market is getting crowded. Mistral’s Leanstral 1.5 for code verification, Meta’s Muse Spark 1.1 for autonomous agents, and Moonshot’s upcoming Kimi K3 (2.8 trillion parameters, releasing July 27) all target overlapping but distinct use cases.
FAQ
What is Thinking Machines Lab’s Inkling model?
Inkling is a 975-billion-parameter open-weight AI model from Thinking Machines Lab, founded by former OpenAI CTO Mira Murati. It supports text, image, and audio inputs and is designed for enterprise customization.
How does Inkling compare to GPT-5.6 and Claude?
Inkling is smaller and not positioned as a direct competitor to frontier models like GPT 5.6 Sol or Claude Fable 5. Instead, it focuses on customizability and on-premises deployment for enterprises.
Can developers use Inkling for free?
Yes. The model weights are released openly, so developers can download, modify, and deploy Inkling without licensing fees. The training code and data remain proprietary.
Who funded Thinking Machines Lab?
Thinking Machines raised a $2 billion seed round at a $12 billion valuation in 2025, with Nvidia as a key investor. The company also reportedly signed a multibillion-dollar Google Cloud deal.
What is Mira Murati’s background?
Mira Murati served as CTO of OpenAI, where she oversaw the release of GPT-4 and led the DALL-E project. She left OpenAI in late 2024 to found Thinking Machines Lab.
