Mira Murati, the former Chief Technology Officer of OpenAI, has announced the first product from her new artificial intelligence startup, Thinking Machines. The product, called Tinker, is designed to make it easier for developers, researchers, and businesses to fine-tune large language models using their own laptops. This marks a notable shift in the accessibility of advanced AI tools, as the process of training or customizing large models has traditionally been limited to those with access to expensive and powerful computing infrastructure.
Murati, who played a central role in scaling OpenAI’s research and product portfolio before leaving the company earlier this year, said the vision behind Thinking Machines is to democratize access to AI. With Tinker, her team aims to give individuals and smaller organizations the ability to adapt models to their own datasets and use cases without relying entirely on cloud providers or advanced data centers. By removing barriers related to cost and hardware, the company hopes to empower a wider base of users to participate in the AI ecosystem.
The product launch comes at a time when fine-tuning large language models has become a critical need across industries. Businesses are increasingly seeking to customize foundational models to reflect their brand tone, domain-specific vocabulary, or regulatory requirements. However, this has often required renting expensive GPU clusters in the cloud, creating both technical and financial hurdles. Thinking Machines is positioning Tinker as a tool that reduces these hurdles by allowing fine-tuning to run on locally available hardware, opening up opportunities for broader adoption.
Industry observers note that Murati’s entry into this segment is significant, given her background and credibility in the field. During her time at OpenAI, she was instrumental in the development and launch of widely used systems like ChatGPT and DALL-E, which helped set the stage for AI’s mainstream adoption. With Thinking Machines, she is pivoting toward a more user-focused approach, placing control and customization in the hands of developers rather than concentrating power within a few large AI labs.
According to early demonstrations of Tinker, the platform allows users to input their own datasets and apply lightweight fine-tuning methods that do not require excessive processing power. While full-scale training of massive models is still out of reach for individual machines, the tool relies on techniques like parameter-efficient fine-tuning and quantization to enable effective adjustments. This means companies can adapt models for specialized purposes, such as creating legal assistants trained on case law, healthcare bots aligned with medical guidelines, or retail systems attuned to customer data, all while retaining control of sensitive information.
Murati highlighted privacy as one of the drivers behind the product design. In an era where businesses and governments are becoming increasingly cautious about data sharing, Tinker offers a way for organizations to fine-tune models locally, without exposing proprietary or sensitive datasets to third-party servers. This aligns with a broader trend in the AI industry toward more decentralized and privacy-conscious workflows.
Reactions to the launch have been largely positive, with experts pointing out that a tool like Tinker could help bridge the gap between cutting-edge AI research and practical, everyday applications. Startups and smaller firms, in particular, stand to benefit from being able to fine-tune models without the steep overhead that has historically accompanied such work.
However, challenges remain. Running AI training on personal laptops will inevitably come with limitations in scale and performance. Tinker is unlikely to replace large-scale cloud operations for enterprise-level model development. Instead, its value proposition lies in enabling lightweight customization and experimentation, which can then be scaled up as needed. This positions the product as an entry point rather than a replacement for cloud-based solutions.
The launch also highlights the increasingly competitive landscape of AI tooling. With giants like Microsoft, Google, and Amazon continuing to integrate customization options into their platforms, Thinking Machines is carving a niche by focusing on accessibility and simplicity. Murati’s leadership and track record provide an additional boost, lending credibility to a startup that is likely to attract attention from both developers and investors.
As businesses worldwide grapple with how to make AI systems align more closely with their goals, the importance of accessible fine-tuning tools will continue to grow. By allowing these processes to happen on devices as common as laptops, Thinking Machines is signaling a future where the power to shape AI is distributed more widely.
For Murati, Tinker represents more than just a product launch. It is also a statement about the future of AI innovation — one where personalization, accessibility, and user control play central roles. If the product succeeds in lowering the barriers to entry, it could spark a wave of new applications and creative uses of AI across industries that have so far been hesitant to adopt it.
The debut of Tinker underscores the shift toward making AI less of a black box and more of a tool that organizations can directly shape to their needs. In doing so, Murati and her team at Thinking Machines are pushing the industry in a direction where the ability to fine-tune models is no longer a privilege reserved for large technology companies but an accessible capability for anyone with a dataset and a laptop.