Thinking Machines Lab Unveils Tinker: A New API for Streamlined LLM Fine-Tuning
Former OpenAI CTO Mira Murati‘s new venture, Thinking Machines Lab, has officially entered the market with it’s debut product: Tinker, a powerful API designed to simplify the process of fine-tuning large language models (LLMs). This release marks a notable step toward democratizing access to advanced AI customization.
Currently in private beta, tinker allows AI and security researchers to experiment with and refine cutting-edge models using Python, all without the complexities of distributed training. Interested organizations can request access directly from Thinking Machines Lab. While currently free, Tinker will transition to a usage-based pricing model in the coming weeks.
What Problems Does Tinker Solve?
Fine-tuning LLMs can be resource-intensive and technically challenging. Tinker addresses these hurdles by offering a streamlined experience.Here’s how:
* Accessibility: It removes the need for complex distributed training setups.
* Cost-Effectiveness: Leveraging low-rank adaptation (LoRa), Tinker optimizes compute usage, reducing overall costs. This allows the same processing power to be shared across multiple training runs.
* Model Variety: Tinker currently supports popular models like Alibaba’s Qwen-235B-A22B and Meta’s Llama-3.2-1B, with plans to expand compatibility.
“Tinker advances our mission of enabling more people to do research on cutting-edge models and customize them to their needs,” the company stated in a recent blog post.
Early adopters already include teams from prestigious institutions like Princeton University, Stanford University, and the University of California, Berkeley, as well as the AI security-focused nonprofit, Redwood Research.
Introducing the Tinker Cookbook
To further empower users,Thinking Machines Lab has released the Tinker Cookbook. This open-source library provides practical examples and pre-built abstractions to accelerate your fine-tuning projects.
The Cookbook offers guidance on:
* Improving math reasoning capabilities.
* Customizing conversational datasets.
* Integrating retrieval tools.
* Optimizing prompt distillation techniques.
Essentially, its a toolkit designed to help you quickly achieve your specific goals within the Tinker habitat.
The Expertise Behind Thinking machines Lab
Thinking Machines Lab isn’t just another AI startup. It’s built by a team of seasoned veterans. Founded in February 2025 by Mira Murati, the lab also includes OpenAI co-founder John Schulman and former employees from leading AI organizations like Google, Meta, Mistral, and Character AI.
This deep industry experience is already evident in the quality of their work. In September, the lab published research focused on reducing nondeterminism in generative AI – the frustrating tendency for models to provide inconsistent answers to the same query.
Tinker represents a significant contribution to the evolving landscape of LLM development,offering a powerful and accessible tool for researchers and developers alike. It’s a clear indication that Thinking Machines Lab is poised to become a key player in the future of artificial intelligence.
