Thinking Machines Lab, a heavily funded startup cofounded by prominent researchers from OpenAI, has revealed its first product—a tool called Tinker that automates the creation of custom frontier AI models.
“We believe [Tinker] will help empower researchers and developers to experiment with models, and will make frontier capabilities much more accessible to all people,” says Mira Murati, cofounder and CEO of Thinking Machines, in an interview with WIRED ahead of the announcement.
Big companies and academic labs already fine-tune open source AI models to create new variants that are optimized for specific tasks, like solving math problems, drafting legal agreements, or answering medical questions.
Typically, this work involves acquiring and managing clusters of GPUs and using various software tools to ensure that large-scale training runs are stable and efficient. Tinker promises to allow more businesses, researchers, and even hobbyists to fine-tune their own AI models by automating much of this work.
Essentially, the team is betting that helping people fine-tune frontier models will be the next big thing in AI. And there’s reason to believe they might be right. Thinking Machines Lab is helmed by researchers who played a core role in the creation of ChatGPT. And, compared to similar tools on the market, Tinker is more powerful and user friendly, according to beta testers I spoke with.
Murati says that Thinking Machines Lab hopes to demystify the work involved in tuning the world’s most powerful AI models, and make it possible for more people to explore the outer limits of AI. “We’re making what is otherwise a frontier capability accessible to all, and that is completely game changing,” she says. “There are a ton of smart people out there, and we need as many smart people as possible to do frontier AI research.”
Tinker currently allows users to fine-tune two open source models: Meta’s Llama and Alibaba’s Qwen. Users can write a few lines of code to tap into the Tinker API and start fine-tuning through supervised learning, which means adjusting the model with labeled data, or through reinforcement learning, an increasingly popular method for tuning models by giving them positive or negative feedback based on their outputs. Users can then download their fine-tuned model and run it wherever they want.
The AI industry is watching the launch closely—in part due to the caliber of the team behind it.
Murati was previously the CTO of OpenAI. She briefly served as OpenAI’s CEO after the board ousted Sam Altman in late 2023. Roughly 10 months later, she announced that she was leaving the firm.
Murati cofounded Thinking Machines Lab with a handful of other OpenAI veterans, including John Schulman, an OpenAI cofounder; Barret Zoph, the ex-vice president of research; Lilian Weng, who worked on safety and robotics research; Andrew Tulloch, who worked on pretraining and reasoning; and Luke Metz, a post-training specialist. The team attracted a lot of attention before even announcing any products: In July, the startup revealed that it had raised a whopping $2 billion in seed funding, giving the venture a heady valuation of $12 billion.
Schulman led work on fine tuning the large language model that powers ChatGPT through reinforcement learning. Input from human testers provides a reward signal that makes the model much better (although not perfect) at holding coherent conversations, answering questions without going off track, and avoiding undesirable behaviors. He claims Tinker will make it easier for more people to coax new abilities out of big models by leveraging reinforcement learning and other training tricks. “There’s a bunch of secret magic, but we give people full control over the training loop,” Schulman tells WIRED. “We abstract away the distributed training details, but we still give people full control over the data and the algorithms.”
Thinking Machines Lab will let users apply for access to Tinker starting on Wednesday. The company is not charging for its API for now, though it expects to start doing so eventually.
The API has been made available to some beta users already, including Eric Gan, a researcher at Redwood Research, a company focused on the risks posed by AI models, who says he is using Tinker’s reinforcement learning functionality to tune models to write backdoors in code.
Gan says Tinker has made it possible to coax capabilities out of a model that simply wouldn’t be revealed using an API. It is relatively easy to make adjustments to the fine-tuning, he says. “Tinker is definitely much simpler than doing the RL from scratch,” Gan notes, adding: “RL is especially good for if you have a very specialized task and existing models aren’t capable of doing it.”
Another beta tester, Robert Nishihara, the CEO of Anyscale, a company that supplies technology for managing large-scale AI projects, says that while other fine-tuning tools like VERL and SkyRL already exist, Tinker offers a remarkable mix of abstraction and tunability. “I think it’s a great API and a lot of people will want to use it,” he says.
A lingering fear around open source models is that they can be downloaded and modified in nefarious ways. Thinking Machines currently vets those who get access to its API but Schulman says the company will eventually introduce automated systems to guard against misuse.
Tinker might be Thinking Machines Lab’s first product, but the company has already been publishing fundamental research on model training, including advances on maintaining the performance of neural networks and fine-tuning large language models more efficiently, which it leverages behind the scenes for tools like Tinker.
The company’s plan to open up the process of tuning big models also shows a commitment to openness at a time when most US AI companies keep their best models closed and accessible only through an API. China currently has more open source frontier AI models than the US, and these models are being used by many companies and researchers around the world.
Murati says she hopes that Tinker will help reverse the trend of commercial AI models becoming increasingly closed. “If you consider what’s being done in frontier labs and what other smart people in the world of academia, they’re sort of diverging more and more,” she says. “And that’s not great if you think about how these powerful systems are coming into the world.”
