Mira Murati's Thinking Machines Lab Launches Tinker, Democratizing AI Fine-Tuning for Everyone

The AI landscape just shifted. Mira Murati—former CTO of OpenAI and one of the most influential figures in artificial intelligence—has unveiled Tinker, the first product from her $12 billion startup Thinking Machines Lab. It's a bet that the future of AI isn't about building bigger models. It's about letting everyone build better ones.

What is Tinker?

Tinker is a flexible API for fine-tuning language models. In plain terms: it lets researchers, developers, and hackers customize AI models to their specific needs without needing Google-scale infrastructure.

The pitch is compelling: 90% algorithmic control with 90% less infrastructure complexity.

You can fine-tune models ranging from small open-weight options to massive mixture-of-experts architectures like Qwen-235B. Switching between model sizes is as simple as changing a parameter.

"We believe [Tinker] will help empower researchers to experiment with models by giving them control over the algorithms and data while we handle the complexity of distributed training."

— Thinking Machines Lab

Why This Matters

Fine-tuning has always been the bottleneck. You could access GPT-4 or Claude through an API, but actually customizing those models to your domain? That required:

• Millions in compute
• A dedicated ML ops team
• Access to proprietary training infrastructure

Tinker removes those barriers. With just four core functions—forward_backward, optim_step, sample, and save—researchers can run sophisticated training loops on frontier-scale models.

Early Results Are Promising

Researchers at Stanford, Berkeley, and Princeton are already using Tinker to push the boundaries in reinforcement learning, computational chemistry, and theorem proving. The Tinker Cookbook provides ready-to-run examples for common use cases.

Open Science, Open Access

In a move that signals Thinking Machines' philosophical stance, the company is offering research and teaching grants:

• $250 in free credits per student for academic courses
• Research grants for scholars working on open-weight LLMs

This isn't charity—it's strategy. By making Tinker the default tool for AI research, Thinking Machines is positioning itself at the center of the next wave of AI innovation.

The Bigger Picture

Murati's departure from OpenAI in September 2024 raised eyebrows. Now, with Thinking Machines valued at $12 billion before shipping a product, her vision is becoming clear:

The first superintelligence won't be a superhuman reasoner. It'll be a superhuman learner.

That's the thesis driving Thinking Machines. While OpenAI, Anthropic, and Google race to build bigger models, Murati is betting on the tools that let everyone else build smarter ones.

Rafael Rafailov, a reinforcement learning researcher at Thinking Machines, put it bluntly at TED AI: "The path forward isn't about training bigger—it's about learning better."

What's Next

Tinker is just the beginning. Thinking Machines has hinted at more products with "significant open source components." For a company founded by ex-OpenAI researchers who watched the transition from "open" to "closed," that's a statement of intent.

The AI wars are heating up. And Thinking Machines just entered the arena with a fundamentally different playbook.

Tinker is available now at thinkingmachines.ai/tinker. Research grant applications are open.