Scholar Feed

LLM reasoning / chain-of-thought

ORM

Heavily superseded#4 of 772 most-superseded

Heavily superseded — a standard baseline that newer methods routinely beat

10 papers critique it · 4 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites ORM as a baseline.

Early alignment efforts relied on Outcome Reward Models (ORM)… where feedback is concentrated solely on the final solution; however, this approach faces a severe credit assignment problem
— CoTZero: Annotation-Free Human-Like Vision Reasoning via Hierarchical Synthetic CoT
“Process Reward Models (PRMs) enable fine-grained step-level supervision for model reasoning, addressing traditional Outcome Reward Models (ORMs) limitation of only scoring final outputs.”
— CoLD: Counterfactually-Guided Length Debiasing for Process Reward Models
“This granularity mismatch leads to inconsistent scoring between partial and complete sequences.”
— From Outcomes to Processes: Guiding PRM Learning from ORM for Inference-Time Alignment
“Existing works wang2024mathshepherdverifyreinforcellms, lightman2023letsverifystepstep show that PRM outperforms ORM by a considerable margin”
— Entropy-Regularized Process Reward Model
“most of recent work demonstrates that ORMs fall short on complex multi-step reasoning tasks”
— Demystifying Multilingual Chain-of-Thought in Process Reward Modeling
“ORM evaluates the whole reasoning process based on the final answer, ignoring intermediate steps”
— AURORA:Automated Training Framework of Universal Process Reward Models via Ensemble Prompting and Reverse Verification
“Traditional Outcome Reward Models (ORMs)...which assign rewards based solely on the final answer, fail to detect flawed intermediate reasoning.”
— MM-PRM: Enhancing Multimodal Mathematical Reasoning with Scalable Step-Level Supervision
“Outcome Reward Models (ORMs) focus solely on evaluating final solutions for correctness, ignoring process optimality and therefore missing costly but non-terminal inefficiencies”
— When Agents go Astray: Course-Correcting SWE Agents with PRMs
“ORM evaluates the entire reasoning path by assigning a single score to the final solution, whereas PRM provides step-level scores, yielding denser reward signals and generally outperforming ORM.”
— Exploring Generative Process Reward Modeling for Semi-Structured Data: A Case Study of Table Question Answering
“this approach, which provides a sparse reward risks inadvertently validating reasoning trajectories that are flawed, illogical, or factually incorrect, so long as they coincidentally arrive at the correct final output”
— DC-W2S: Dual-Consensus Weak-to-Strong Training for Reliable Process Reward Modeling in Biological Reasoning

Beaten on benchmarks

Head-to-head results where a newer method reports beating ORM. Values are copied from the source paper's tables — verify against the cited paper.

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.