Thinking Machines Lab Unveils Tinker: Revolutionizing AI Model Fine-Tuning

Reviewed byNidhi Govil

4 Sources

Share

Mira Murati's AI startup launches Tinker, a powerful API for customizing language models, aiming to democratize frontier AI capabilities.

Thinking Machines Lab Introduces Tinker

Thinking Machines Lab, a heavily funded AI startup cofounded by former OpenAI CTO Mira Murati, has unveiled its first product - Tinker, an API designed to automate and simplify the creation of custom frontier AI models

1

2

. This launch marks a significant milestone for the company, which raised $2 billion earlier this year at a $12 billion valuation

3

.

Source: SiliconANGLE

Source: SiliconANGLE

Democratizing AI Model Fine-Tuning

Tinker aims to make frontier AI capabilities more accessible to researchers, developers, and even hobbyists. The tool automates much of the complex work involved in fine-tuning large language models (LLMs), including managing GPU clusters and ensuring stable, efficient training runs

1

.

Murati emphasizes the transformative potential of Tinker: "We're making what is otherwise a frontier capability accessible to all, and that is completely game changing"

1

.

Source: Silicon Republic

Source: Silicon Republic

Technical Capabilities and Features

Tinker offers a Python-based API that provides granular control over the fine-tuning process while handling the complexities of distributed training

2

. Key features include:

  1. Support for both small and large open-weight models, including Mixture-of-Experts architectures like Qwen-235B-A22B
  2. Integration with LoRA (Low-Rank Adaptation) for cost-efficient training
  3. An open-source companion library called the Tinker Cookbook
  4. Flexibility to fine-tune models through supervised learning or reinforcement learning

    1

    2

Source: VentureBeat

Source: VentureBeat

Early Adoption and Use Cases

Tinker has already been adopted by several research institutions, demonstrating its versatility across different domains:

  • Princeton's Goedel Team used Tinker for fine-tuning LLMs in formal theorem proving
  • Stanford's Rotskoff Lab applied it to train chemical reasoning models
  • Berkeley's SkyRL implemented custom multi-agent reinforcement learning loops
  • Redwood Research utilized Tinker for RL-training on long-context AI control tasks

Industry Reception and Potential Impact

The AI community has responded positively to Tinker's launch. Andrej Karpathy, former OpenAI co-founder, praised Tinker's design tradeoffs, noting that it allows users to retain "~90% of algorithmic control while removing ~90% of infrastructure pain" .

John Schulman, chief scientist and co-founder of Thinking Machines, described Tinker as "the infrastructure I've always wanted" .

Availability and Pricing

Tinker is currently available in private beta, with Thinking Machines starting to onboard users immediately. While initially free to use, the company plans to introduce usage-based pricing in the coming weeks

3

4

.

As the AI industry closely watches this launch, Tinker's potential to democratize frontier AI capabilities could significantly impact the landscape of AI research and development, enabling a broader range of individuals and organizations to contribute to and benefit from advancements in language models.

TheOutpost.ai

Your Daily Dose of Curated AI News

Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Spend less time searching for the latest in AI and get straight to action.

© 2025 Triveous Technologies Private Limited
Instagram logo
LinkedIn logo