# Reward Modeling on JudgeBench (Positional Consistency) Source: [Reward Modeling on JudgeBench (Positional Consistency)]() Collection index: [llms.txt]() Data updated: 2026-04-10T07:20:53.251Z Evaluation of model performance on the JudgeBench dataset for the Reward Modeling task, specifically reporting the Positional Consistency score. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Reward Modeling | | Dataset (including any stated split) | JudgeBench | | Metric | Positional Consistency Score | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 8 of 8 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Positional Consistency Score](): Higher is better; unit: PERCENT. ## Recorded results | Method | Positional Consistency Score | Result date | Report paper | | --- | --- | --- | --- | | ReflectRM | 56.3 | Not recorded | 69d883e07d0172140ec77804 | | ReflectRM | 54.9 | Not recorded | 69d883e07d0172140ec77804 | | Unified Training | 53.9 | Not recorded | 69d883e07d0172140ec77804 | | RFT | 53.1 | Not recorded | 69d883e07d0172140ec77804 | | Unified Training | 53.1 | Not recorded | 69d883e07d0172140ec77804 | | Qwen3-8B | 52.7 | Not recorded | 69d883e07d0172140ec77804 | | Qwen3-4B | 49.3 | Not recorded | 69d883e07d0172140ec77804 | | RFT | 48.9 | Not recorded | 69d883e07d0172140ec77804 | ### ReflectRM Positional Consistency Score: 56.3. Result ID: 69d8a4ba47f34ce924c11a7f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Training Method":"Unified Training","Inference Strategy":"Two-Stage Inference"}. This record does not assert independent reproduction. ### ReflectRM Positional Consistency Score: 54.9. Result ID: 69d8a4ba47f34ce924c11a7b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B","Training Method":"Unified Training","Inference Strategy":"Two-Stage Inference"}. This record does not assert independent reproduction. ### Unified Training Positional Consistency Score: 53.9. Result ID: 69d8a4ba47f34ce924c11a7e. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Training Method":"Unified Training"}. This record does not assert independent reproduction. ### RFT Positional Consistency Score: 53.1. Result ID: 69d8a4ba47f34ce924c11a79. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B","Training Method":"RFT"}. This record does not assert independent reproduction. ### Unified Training Positional Consistency Score: 53.1. Result ID: 69d8a4ba47f34ce924c11a7a. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B","Training Method":"Unified Training"}. This record does not assert independent reproduction. ### Qwen3-8B Positional Consistency Score: 52.7. Result ID: 69d8a4ba47f34ce924c11a7c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B"}. This record does not assert independent reproduction. ### Qwen3-4B Positional Consistency Score: 49.3. Result ID: 69d8a4ba47f34ce924c11a78. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B"}. This record does not assert independent reproduction. ### RFT Positional Consistency Score: 48.9. Result ID: 69d8a4ba47f34ce924c11a7d. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Training Method":"RFT"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.