What is Reinforcement Fine-Tuning?
Reinforcement Fine-Tuning is a new model customization technique that enables customers to create “expert models” for a narrow set of tasks in their domain. It allows for:
Learning from user-provided inputs and a grader to evaluate model outputs.
Iterative improvement, optimization, and customization of the model’s chain-of-thought to perform best on the provided tasks.
Reinforcement Fine-Tuning is becoming available to a small group of alpha users as part of our research preview program.
In this alpha phase, we are working with select participants to gather feedback on how this feature can be used effectively and responsibly. At this time, we do not have a timeline or additional details to share on broader public availability.
We are actively gathering insights from alpha users to refine this feature. For updates, keep an eye on our Twitter and our website.
For more information and requesting access, please refer to