Reinforcement Fine-Tuning (removed)

Managed reinforcement fine-tuning has been removed. Find your existing RFT jobs with firectl, and run new reinforcement learning on the Training API.

Managed reinforcement fine-tuning has been removed. It accepts no new jobs, and existing jobs can’t run or resume. Reinforcement learning runs on the Training API. Start with Cookbook: Reinforcement Learning.

Cookbook: Reinforcement Learning

Async GRPO with a custom rollout function on the Training API.

Cookbook: Agentic RL

Multi-turn harnesses, tools, sandboxes, and token ancestry.

​

Existing jobs

Your managed RFT jobs stay in your account until you delete them. A job that had not finished was marked failed, with a message that managed RFT was removed.

Task	How
List your jobs	firectl reinforcement-fine-tuning-job list or the dashboard
Read one job	firectl reinforcement-fine-tuning-job get or the dashboard
Delete a job	firectl reinforcement-fine-tuning-job delete

The REST API and the Python SDK no longer serve managed RFT jobs.

Models your RFT jobs produced are ordinary models in your account. Deploy them like any other fine-tuned model.

​

Evaluators and evaluation jobs

Evaluators and evaluation jobs have been removed too. On the Training API, your rollout code computes the reward for each sample.