# What is Outcome Reward Model? Definition & Technical Mechanics

> **Source:** [SPIDITS AI AI Glossary](https://spidits.com/ai-glossary)  
> **Category:** Model Training  
> **Canonical Citation:** [https://spidits.com/ai-glossary/outcome-reward-model](https://spidits.com/ai-glossary/outcome-reward-model)

## Definition
An Outcome Reward Model (ORM) is a feedback mechanism that scores only the final response generated by a model, without evaluating the correctness of intermediate reasoning steps. It is simpler to train but less granular than step-by-step reward models.

## Architectural Mechanics & Deep Dive
Outcome Reward Models (ORMs) are feedback systems that score only the final correctness or quality of a model's complete response. While ORMs are easy to configure and require less annotation effort than process-based models, they provide less guidance during multi-step reasoning. Without step-level feedback, ORMs can inadvertently reward model outputs that reach correct conclusions through incorrect or hallucinated logical steps.

## Industry Applications & Real-World Use Cases
Basic classification, text summarization, and simple question-answering validation

## Frequently Asked Questions

### Why would you use an ORM instead of a PRM?
ORMs are much easier and cheaper to train because labeling only the final correctness of a response is faster than labeling every reasoning step.

### What is the risk of using only an ORM for reasoning models?
It can reward "logical alignment by coincidence" where a model arrives at the correct answer through flawed logic or guessing.

## Related Terminology
- [process-reward-model](https://spidits.com/ai-glossary/process-reward-model)
- [reinforcement-learning](https://spidits.com/ai-glossary/reinforcement-learning)
- [chain-of-thought](https://spidits.com/ai-glossary/chain-of-thought)

---
*From the SPIDITS AI Knowledge Base (300+ terms indexed). Explore full technical definitions: [https://spidits.com/ai-glossary](https://spidits.com/ai-glossary)*