Thinking Machines, founded by former OpenAI CTO Mira Murati, has launched Tinker, a Python-based API for large language model (LLM) fine-tuning. The API is now available in a private beta and is free to use.
TL;DR
- Thinking Machines launches Tinker, a Python API for LLM fine-tuning, in private beta.
- Tinker offers developers control over training pipelines while handling distributed training complexity.
- Early adopters include research teams from Princeton, Stanford, Berkeley, and Redwood Research.
What happened
Thinking Machines, an AI startup founded by former OpenAI CTO Mira Murati, has launched its first product, Tinker, a Python-based API designed for LLM fine-tuning. The API is currently in private beta and is free to use.
Tinker provides developers with control over training pipelines, including loss functions, training loops, and data workflows. The actual training workloads run on Thinking Machines' managed infrastructure, enabling fast distributed execution.
Early adopters of Tinker include research teams from Princeton, Stanford, Berkeley, and Redwood Research. These teams have used Tinker for various applications, such as fine-tuning LLMs for formal theorem proving, training chemical reasoning models, and running custom multi-agent reinforcement learning loops.
Why it matters
Tinker's launch is significant for developers and researchers who require granular control over LLM fine-tuning. The API's Python-native primitives and support for both small and large open-weight models make it a versatile tool for various research applications.
The API's ability to handle distributed training complexity while providing a user-friendly interface can significantly reduce the barrier to entry for researchers and developers looking to fine-tune LLMs.
Tinker's free availability during the private beta period and its upcoming pay-as-you-go pricing model make it an attractive option for developers and research teams looking for cost-effective solutions for LLM fine-tuning.
Key facts
- Tinker is a Python-based API for LLM fine-tuning, launched by Thinking Machines.
- The API is currently in private beta and is free to use.
- Tinker offers Python-native primitives for building custom fine-tuning or RL algorithms.
- The API supports both small and large open-weight models, including Mixture-of-Experts architectures like Qwen-235B-A22B.
- Early adopters of Tinker include research teams from Princeton, Stanford, Berkeley, and Redwood Research.
- Tinker's open-source companion library, the Tinker Cookbook, includes implementations of post-training methods.
- Thinking Machines raised $2 billion in funding earlier this year from investors including a16z, NVIDIA, and Accel.
- Tinker is positioned as a developer-centric tool for LLM fine-tuning, offering control over training pipelines while handling distributed training complexity.
Context
Thinking Machines was founded by Mira Murati, who served as CTO of OpenAI until her departure in September 2024. The company's vision emphasizes helping people adapt AI systems to their specific needs, building strong foundations for capable and safe AI, and fostering open science through public releases of models, code, and research.
The launch of Tinker comes at a time when the open-source AI market has become intensely competitive, with numerous companies fielding powerful models that rival the performance of well-capitalized U.S. labs like OpenAI, Anthropic, Google, and Meta.
As competition for developer mindshare heats up, Thinking Machines is signaling that it's ready to meet demand with a product, technical clarity, and public documentation.