Tinker: A Managed, Low-Level Fine-Tuning API for Open-Weight LLMs

Thinking Machines Lab launched Tinker, a managed fine-tuning platform and API that gives researchers low-level control while abstracting distributed training. It supports everything from small models to large MoE models like Qwen-235B-A22B, uses LoRA to reduce costs, and ships with an open-source Tinker Cookbook of post-training methods. Already used by multiple university and research groups, Tinker is entering private beta today, free to start with usage-based pricing coming soon.

Key Points

• Tinker is a flexible, low-level API for fine-tuning LLMs that exposes primitives like forward_backward and sample.
• It supports a wide range of open-weight models—from small models to very large MoE models like Qwen-235B-A22B—with easy model switching.
• The service is fully managed on internal clusters, handling scheduling, resource allocation, and failure recovery, and uses LoRA to share compute and lower costs.
• An open-source Tinker Cookbook provides modern, ready-to-run implementations of common post-training methods on top of the API.
• Already used by teams at Princeton, Stanford, Berkeley, and Redwood Research; private beta begins today, free initially with usage-based pricing coming soon.

Sentiment

The overall sentiment is mixed. While there's initial praise from an early adopter and recognition of the team's expertise, a significant portion of the discussion revolves around skepticism regarding the product's strategic focus, its differentiation in the market, and a strong debate about the chosen name "Tinker" due to its association with an ethnic slur.

In Agreement

• An alpha tester from Stanford found Tinker "very useful and also really technically impressive," highlighting its effectiveness as a unified framework that abstracts job management complexity while retaining algorithmic flexibility.
• Some believe the product makes sense given the team's expertise at OpenAI, where they excelled at similar infrastructure and fine-tuning challenges.

Opposed

• Disappointment that a company with many smart scientists is focusing "solely on infra and fine-tuning" rather than broader scientific endeavors.
• Questions about Tinker's unique selling proposition (USP), suggesting that fine-tuning features are already offered by many LLM providers.
• Skepticism about the announcement, perceiving it as "yet another announcement of an announcement" without sufficient immediate detail.
• Significant concern raised about the chosen name "Tinker," as it is identified as an antiquated ethnic slur against Irish and Scottish Travellers and Romani people in parts of the Anglosphere, with some arguing this makes it unsuitable for an international product despite its common verb meaning.