What is Thinking Machines Lab? An overview of the ex-OpenAI startup

It’s not every day that a new AI startup, run by the same people who built tools like ChatGPT, manages to raise a jaw-dropping $2 billion in seed funding. Especially when they haven't even launched a product yet. When that happens, you pay attention. The startup in question is Thinking Machine Lab.

Led by former OpenAI CTO Mira Murati and a dream team of AI researchers, the company has landed a staggering $12 billion valuation and is generating the kind of buzz you usually see for a new Marvel movie. But once you get past the hype and the eye-watering numbers, what is Thinking Machine Lab actually trying to build? Let’s cut through the noise and get a clear look at the company, its unique approach, its first product, and what it all means for the future of AI.

What is Thinking Machine Lab?

Founded in February 2025, Thinking Machine Lab is an AI research and product company that got its start when a big chunk of talent left OpenAI. Their mission is to fix some of the biggest problems in AI today by making powerful systems easier to understand, customize, and use for a wider range of builders and researchers. They're all about open science and collaboration, which is a big change from the closed-off, secretive approach we're seeing from many of the industry giants.

This bold vision quickly attracted one of the largest seed rounds in venture capital history. The $2 billion investment was led by Andreessen Horowitz and included a who's who of tech titans like NVIDIA, AMD, and Cisco. That kind of money isn't just a vote of confidence in the all-star team; it's a massive bet on their completely different way of thinking about AI.

With that kind of cash, Thinking Machine Lab isn’t just another company tinkering with AI. It’s setting itself up as a direct challenger to the big players like OpenAI, Google DeepMind, and Anthropic. But they aren't just trying to build a bigger version of what already exists. They're trying to change the rules of the game.

The new Thinking Machine Lab philosophy: Focusing on learning, not just scaling

For the last few years, the main strategy in AI has been pretty simple: just go bigger. The prevailing wisdom was that if you had enough data, enough computing power, and a massive enough model, you could brute-force your way to artificial general intelligence (AGI). Thinking Machine Lab is here to say, "Not so fast."

According to company researcher Rafael Rafailov, the goal isn't just about creating "god-level reasoners." It's about building "superhuman learners." He points out a major flaw in today's best AI systems: they don't really learn from their experiences. You can spend a whole afternoon teaching a a coding assistant how to solve a tricky problem, but when you come back the next day, it's starting from zero again. As Rafailov puts it, for most AI, "every day is their first day of the job."

That approach is incredibly wasteful. Instead of just throwing more data and compute at a problem, Thinking Machine Lab is focused on "meta-learning," which is basically teaching an AI how to learn. The goal is to build systems that can actually remember information, build on past interactions, and get better over time, just like a person does. It's a subtle but powerful shift from training an AI on what to think to giving it the ability to learn how to think for itself. That idea is at the core of everything they're doing.

From research to reality: Introducing Tinker, the first product from Thinking Machine Lab

So, how does this new philosophy turn into an actual product? The company's first offering is a tool called Tinker, and it gives us a pretty clear window into their strategy.

Tinker is an API and a set of tools designed to make it much easier for developers and researchers to customize powerful open-source AI models, like Meta’s Llama. This process is called fine-tuning, where you take a general model and train it to become an expert in a specific task, whether that’s writing legal contracts or answering complex medical questions.

Until now, fine-tuning has been a huge headache. It was expensive and complicated, needing specialized knowledge, tons of GPUs, and fancy software to pull it off. Tinker handles a lot of that heavy lifting. It offers a simple interface that lets users tweak models with just a few lines of code, effectively opening the door for more people to get involved in high-level AI research.

This is a big deal because it empowers innovators who aren't at the big tech labs. Instead of being stuck with the one-size-fits-all APIs from a handful of companies, more people can now experiment with and build their own specialized AI. It’s a first step toward making the tools for building next-gen AI available to everyone.

How your business can use these same ideas today

While Thinking Machine Lab is building for the world's top AI researchers, you don't need a PhD or a billion-dollar valuation to bring the same ideas of customization and specialized learning into your own business. The real magic of AI happens when it's tailored to your specific needs, and you can start doing that right now.

Just as Tinker helps a researcher fine-tune a model for a specific scientific problem, businesses need tools that can specialize AI for their own workflows, especially for things like customer support.

This is exactly where a platform like eesel AI fits in. It’s built on the same core principles but is designed for business teams, not AI scientists.

• Customization based on your data. You don't have to train a massive language model from the ground up. eesel AI gets smart by learning directly from your company's existing knowledge. It connects to your past support tickets, help center articles, and internal documents to understand your brand voice and the real issues your customers face.

• Accessible for the whole team. Tinker opens up AI for developers, but eesel AI makes it accessible for everyone else. It’s a self-serve platform with one-click integrations for tools you already use, like Zendesk, Freshdesk, and Slack. You can have a powerful, custom AI agent up and running in minutes, with no engineers needed.

• Full control over how it works. The control Tinker gives researchers over the training process is similar to the control eesel AI gives support managers. You get to decide exactly which tickets the AI handles, create custom actions for it (like looking up order info in Shopify), and even shape its personality and tone. This makes sure the AI works as a true extension of your team.

Thinking Machine Lab pricing

Right now, you can't just go out and buy Tinker. It's only available for free to a select group of beta users, and there's no public pricing info yet. This is pretty typical for an early-stage company that's still deep in research and development. Their goal at the moment is to get the tool into the hands of researchers to gather feedback, not to make money.

While that makes sense for an R&D lab, businesses need to know the costs. To make smart decisions and manage budgets, you need clear and predictable pricing. It’s one of the key differences between an experimental research tool and a platform that's ready for real-world business use.

The future is specialized AI

Thinking Machine Lab is more than just another startup with a lot of funding. It represents a potential shift in how the AI industry works. With its top-tier team and clear vision, it's poised to move the conversation from a brute-force race for size toward a smarter focus on efficient learning and deep customization.

The main takeaway here is pretty clear: the real power of AI isn't in creating a single, giant model that can do everything. It’s in building smaller, more efficient systems that can be easily adapted to specific needs and datasets.

While Thinking Machine Lab is pioneering that frontier for researchers, businesses can, and should, be applying these same ideas today. The tools are already here to build specialized AI that can solve real problems, automate workflows, and make your team more efficient.

Take the next step with specialized AI

You don't need a $2 billion research lab to build a custom AI for your team. With eesel AI, you can create a specialized AI agent that learns from your knowledge and resolves customer issues instantly. You can get started in just a few minutes.

This video provides an overview of Tinker, the first product launched by Mira Murati's new venture, Thinking Machine Lab.