What it measures

Evaluates reward models' sensitivity to subtle factual/code errors vs. resistance to superficial formatting/markdown style bias ($3\times 3$ style-substance matrix).

Reading the measurement

Separate satisfying instructions from factual accuracy, human preference and safe behavior. Inspect strict all-constraint success as well as average criterion credit. Judge choice and adversarial prompt selection can materially change the result.

Published evaluation

Official RM-Bench leaderboard, reported Overall column. Model type and the source score-mismatch flag are retained. Overall is not recomputed from incomplete category scores.

Where this fits in an evaluation suite

Format compliance, preference, truthfulness, and safety require separate measurements. Keep desirable writing style from masking factual errors.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Official RM-Bench leaderboard, reported Overall column. Model type and the source score-mismatch flag are retained. Overall is not recomputed from incomplete category scores.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

  1. RM-Bench: Benchmarking Reward Models of Language Models with Subtlety and Style — published result table ↗

    Liu, Yantao, Yao, Zijun, Min, Rui, Cao, Yixin, Hou, Lei, Li, Juanzi · 2024

Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.