Thinking Machines Lab has put its first in-house AI model into public view. The model, called Inkling, is open-weight, which means outside developers and companies can download it and modify it directly.
That choice places Inkling in a different lane from flagship systems sold by OpenAI, Anthropic, and Google. Instead of presenting one general-purpose chatbot as the main product, Thinking Machines is arguing for AI that organizations can shape around their own work.
Inkling is built for adaptation, not a single fixed use
Inkling is a mixture-of-experts system with 975 billion total parameters. For any given task, it uses about 41 billion of them, a design meant to make a very large model faster and cheaper to run than activating the full system every time.
The model was trained on 45 trillion tokens of text, image, audio, and video. According to the company, it reasons natively across all four formats, although its current outputs are limited to text. Those outputs include code, styled artifacts, and structured data.
Thinking Machines is not presenting Inkling as the strongest model on the market. Its own blog post says Inkling is “not the strongest overall model available today, open or closed.” The message is more measured: the company is aiming for a broad, adaptable system that can become more useful after organizations customize it.
That distinction matters for enterprise AI. A model that is open-weight can become a starting point rather than a sealed service. Developers can inspect it, modify it, and build around it in ways that are harder with closed systems accessed only through metered products.
The company is betting against one-size-fits-all AI
Thinking Machines Lab was founded by former OpenAI CTO Mira Murati. Inkling is the company’s first public proof point after a year and a half spent building AI infrastructure largely outside public view.
Some of that work appeared earlier in a May research preview of “interaction models.” Those systems were described as AI designed to listen, speak, and even interrupt, instead of behaving like typical chatbots that stop and wait between turns.
Inkling extends the same broader argument: useful AI should not always be centrally trained by one company and then treated as finished. In a post published last week, Thinking Machines argued that centrally trained, fixed AI underperforms systems that organizations shape themselves, because much expertise is specific to the people who hold it.
The company’s enterprise pitch is built around that idea. Inkling is being marketed less as a complete, final product and more as a base model that organizations can fine-tune through Tinker, the company’s model-customization platform.
That approach comes with a tradeoff. Customers that customize the model are also responsible for making sure their changes are safe. The source article notes that fine-tuning requires serious machine-learning talent, which means Inkling is not aimed at every business that wants a chatbot.
Cost, speed, and uncertainty are part of the pitch
Inkling includes controls that let users raise or lower “thinking effort” when they want to trade between speed and deeper processing. It is also designed to give calibrated answers, including flagging uncertainty instead of guessing.
Thinking Machines has pointed to efficiency as one of Inkling’s strengths. On one benchmark, the company says Inkling uses a third as many tokens as Nvidia’s Nemotron 3 Ultra to reach the same coding performance.
The broader business argument is also about where enterprise AI costs sit. In a blog post published Sunday, Microsoft CEO Satya Nadella warned that companies using proprietary AI models effectively pay twice: once through subscriptions and again by giving up business knowledge contained in prompts and corrections that may be absorbed into future model versions.
Hugging Face CEO Clem Delangue made a similar prediction in conversation with TechCrunch last week. He said frontier models will increasingly be used for experimentation and high-value tasks, while most production AI work shifts to private or open source alternatives.
Those arguments fit the lane Thinking Machines is trying to occupy. If many companies want AI that reflects private expertise, then an open-weight model paired with customization tools may be more important than a single model that wins every general benchmark.
Bridgewater shows the clearest example so far
The strongest example cited for this strategy is a recent project with Bridgewater Associates, the world’s largest hedge fund. Bridgewater is not a Thinking Machines investor, according to the source article.
Researchers from both companies took an existing open source model and trained it further on Bridgewater’s financial expertise. The result was said to score 84.7% on financial reasoning tests, beating top proprietary AI models, while costing roughly a fourteenth as much to run.
Those results came from the two companies’ own evaluation, not an independent one. Even so, the project illustrates the logic behind Inkling: a customized model may beat a stronger general model inside a specialized domain if it can absorb the organization’s own knowledge.
That is also why Tinker is central to the business. Once model weights are public, anyone who downloads them is not required to pay Thinking Machines to run them. The company’s revenue case depends on training, fine-tuning, and a cut of the hosting ecosystem built around its model-customization platform.
Open questions remain around training and economics
Thinking Machines says it pre-trained Inkling from scratch. It also says it used other open-weight models, including Moonshot AI’s Kimi K2.5, to help generate some early post-training data before large-scale reinforcement learning took over. The company says its next model will use fully self-contained post-training.
The cost picture is less clear. Thinking Machines struck a partnership with Nvidia in March to deploy a gigawatt of Vera Rubin computing capacity, and Inkling was trained entirely on Nvidia’s GB300 NVL72 systems. The company has not said how it plans to cover those costs.
A reported $50 billion fundraising round was said to be coming together last November but had stalled by January. Since then, the company has declined to discuss its funding picture.
Headcount appears more settled. Thinking Machines now employs roughly 200 people, after earlier departures that included two co-founders who left for OpenAI in January.
Inkling therefore arrives as both a model release and a statement of strategy. Thinking Machines is not trying to copy the general-purpose chatbot path taken by OpenAI, Anthropic, and Google. It is testing whether enterprises will pay for the ability to make AI more their own.