# Reward Modeling on RewardBench Chat-Hard

Source: [Reward Modeling on RewardBench Chat-Hard](<https://www.sota2.com/research/sota/reward-modeling-on-rewardbench-chat-hard>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-18T07:22:39.925Z

Evaluation of model performance on the specific Chat-Hard subset of the RewardBench dataset for the Reward Modeling task, measured using the Chat-Hard metric.

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) | RewardBench Chat-Hard |
| Metric | Chat-Hard Score |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| 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

- [Chat-Hard Score](<https://www.sota2.com/research/sota/reward-modeling-on-rewardbench-chat-hard.md?metric=Chat-Hard%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Chat-Hard Score | Result date | Report paper |
| --- | --- | --- | --- |
| SKYWORK-RM (w/ calibration) | 89.4 | Not recorded | 68fa780d1e209131cea768f9 |
| SKYWORK-RM | 88.8 | Not recorded | 68fa780d1e209131cea768f9 |
| SKYWORK-RM (w/o calibration) | 88.8 | Not recorded | 68fa780d1e209131cea768f9 |
| GRM-RM (w/ calibration) | 68.9 | Not recorded | 68fa780d1e209131cea768f9 |
| GRM-RM (w/o calibration) | 68.8 | Not recorded | 68fa780d1e209131cea768f9 |
| GRM-RM | 68.4 | Not recorded | 68fa780d1e209131cea768f9 |
| FSFAIRX-RM (w/ calibration) | 65.7 | Not recorded | 68fa780d1e209131cea768f9 |
| FSFAIRX-RM | 65.3 | Not recorded | 68fa780d1e209131cea768f9 |
| FSFAIRX-RM (w/o calibration) | 65.1 | Not recorded | 68fa780d1e209131cea768f9 |
| MISTRAL-RM (w/ calibration) | 65.1 | Not recorded | 68fa780d1e209131cea768f9 |
| MISTRAL-RM (w/o calibration) | 62.5 | Not recorded | 68fa780d1e209131cea768f9 |
| MISTRAL-RM | 60.5 | Not recorded | 68fa780d1e209131cea768f9 |
| BTRM-RM (w/o calibration) | 58.7 | Not recorded | 68fa780d1e209131cea768f9 |
| BTRM-RM | 58.1 | Not recorded | 68fa780d1e209131cea768f9 |
| BTRM-RM (w/ calibration) | 57.8 | Not recorded | 68fa780d1e209131cea768f9 |

### SKYWORK-RM (w/ calibration)

Chat-Hard Score: 89.4. Result ID: 69ba52a947f34ce924a3ca60.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.639","Calibration setting":"w/ calibration"}. This record does not assert independent reproduction.

### SKYWORK-RM

Chat-Hard Score: 88.8. Result ID: 69ba52a947f34ce924a3ca5e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa755d1e209131cea67806.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.639","Calibration setting":"Original"}. This record does not assert independent reproduction.

### SKYWORK-RM (w/o calibration)

Chat-Hard Score: 88.8. Result ID: 69ba52a947f34ce924a3ca5f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa755d1e209131cea67806.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.639","Calibration setting":"w/o calibration"}. This record does not assert independent reproduction.

### GRM-RM (w/ calibration)

Chat-Hard Score: 68.9. Result ID: 69ba52a947f34ce924a3ca69.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.508","Calibration setting":"w/ calibration"}. This record does not assert independent reproduction.

### GRM-RM (w/o calibration)

Chat-Hard Score: 68.8. Result ID: 69ba52a947f34ce924a3ca68.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75501e209131cea6715f.md>)

[Implementation](<https://github.com/YangRui2015/Generalizable-Reward-Model>)

Recorded evaluation configuration: {"Mismatch Degree":"0.508","Calibration setting":"w/o calibration"}. This record does not assert independent reproduction.

### GRM-RM

Chat-Hard Score: 68.4. Result ID: 69ba52a947f34ce924a3ca67.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75501e209131cea6715f.md>)

[Implementation](<https://github.com/YangRui2015/Generalizable-Reward-Model>)

Recorded evaluation configuration: {"Mismatch Degree":"0.508","Calibration setting":"Original"}. This record does not assert independent reproduction.

### FSFAIRX-RM (w/ calibration)

Chat-Hard Score: 65.7. Result ID: 69ba52a947f34ce924a3ca63.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.554","Calibration setting":"w/ calibration"}. This record does not assert independent reproduction.

### FSFAIRX-RM

Chat-Hard Score: 65.3. Result ID: 69ba52a947f34ce924a3ca61.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa715d1e209131cea50de2.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.554","Calibration setting":"Original"}. This record does not assert independent reproduction.

### FSFAIRX-RM (w/o calibration)

Chat-Hard Score: 65.1. Result ID: 69ba52a947f34ce924a3ca62.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa715d1e209131cea50de2.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.554","Calibration setting":"w/o calibration"}. This record does not assert independent reproduction.

### MISTRAL-RM (w/ calibration)

Chat-Hard Score: 65.1. Result ID: 69ba52a947f34ce924a3ca66.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.528","Calibration setting":"w/ calibration"}. This record does not assert independent reproduction.

### MISTRAL-RM (w/o calibration)

Chat-Hard Score: 62.5. Result ID: 69ba52a947f34ce924a3ca65.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.528","Calibration setting":"w/o calibration"}. This record does not assert independent reproduction.

### MISTRAL-RM

Chat-Hard Score: 60.5. Result ID: 69ba52a947f34ce924a3ca64.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.528","Calibration setting":"Original"}. This record does not assert independent reproduction.

### BTRM-RM (w/o calibration)

Chat-Hard Score: 58.7. Result ID: 69ba52a947f34ce924a3ca6b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.162","Calibration setting":"w/o calibration"}. This record does not assert independent reproduction.

### BTRM-RM

Chat-Hard Score: 58.1. Result ID: 69ba52a947f34ce924a3ca6a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.162","Calibration setting":"Original"}. This record does not assert independent reproduction.

### BTRM-RM (w/ calibration)

Chat-Hard Score: 57.8. Result ID: 69ba52a947f34ce924a3ca6c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa780d1e209131cea768f9.md>)

Recorded evaluation configuration: {"Mismatch Degree":"0.162","Calibration setting":"w/ calibration"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
