AI Glossary

Reinforcement learning from verifiable rewards

Also called RLVR

What does “Reinforcement learning from verifiable rewards” mean?

Reinforcement learning from verifiable rewards: Scores a model's outputs with an automatic check instead of a learned preference model. The check might be whether a math answer is correct or whether code passes its tests. It has become a common way to train reasoning models on tasks with clear right answers.

Reinforcement learning from verifiable rewards visual guide: Scores a model's outputs with an automatic check instead of a learned preference model. The check might be whether a math answer is correct or whether code passes its tests. It has become a common way to train reasoning models on tasks with clear right answers.
Visual guide Open full size (new tab)

What is an example of “Reinforcement learning from verifiable rewards”?

A unit test either passes or fails, so coding tasks give a clear reward signal that is harder for the model to game than a vague quality rating.

Keep learning

See also

Reinforcement learningTermReward modelTermReasoning modelTermReinforcement learning from human feedbackTerm

Same topic

Scaling lawsTermReinforcement learning from AI feedbackTermSynthetic dataTermRegularizationTermTransfer learningTerm