# Reward Model Evaluation on RewardBench (Accuracy)

Source: [Reward Model Evaluation on RewardBench (Accuracy)](<https://www.sota2.com/research/sota/reward-model-evaluation-on-rewardbench-accuracy>)

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

Data updated: 2026-06-01T08:27:48.807Z

Evaluation of model performance on the RewardBench dataset using Accuracy as the primary 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 Model Evaluation |
| Dataset (including any stated split) | RewardBench |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 12 of 12 |
| 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

- [Accuracy](<https://www.sota2.com/research/sota/reward-model-evaluation-on-rewardbench-accuracy.md?metric=Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| SAVE | 93.9 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| SAVE | 93.6 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| SAVE (w/o Curriculum Mechanism) | 93.6 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| SAVE (w/o Policy Model Optimization) | 93.6 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| SAVE (w/o Curriculum Mechanism) | 93.6 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| SAVE (w/o Policy Model Optimization) | 93.5 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| HL-BT | 93.3 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| HL-BT | 93.2 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| Continual Offline Training RM | 93.1 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| Skywork-Reward-V2-Llama-3.2-3B | 93 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| Mean Reward | 83.4 | Not recorded | 6a1d11dfdbd7150506f83dc6 |
| Mean Reward | 83.1 | Not recorded | 6a1d11dfdbd7150506f83dc6 |

### SAVE

Accuracy: 93.9. Result ID: 6a1d4273fbc373592c166ec1.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen3-4B-Instruct-2507","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### SAVE

Accuracy: 93.6. Result ID: 6a1d4273fbc373592c166ebc.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen2.5-3B-Instruct","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### SAVE (w/o Curriculum Mechanism)

Accuracy: 93.6. Result ID: 6a1d4273fbc373592c166ebd.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen2.5-3B-Instruct","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### SAVE (w/o Policy Model Optimization)

Accuracy: 93.6. Result ID: 6a1d4273fbc373592c166ebe.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen2.5-3B-Instruct","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### SAVE (w/o Curriculum Mechanism)

Accuracy: 93.6. Result ID: 6a1d4273fbc373592c166ec2.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen3-4B-Instruct-2507","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### SAVE (w/o Policy Model Optimization)

Accuracy: 93.5. Result ID: 6a1d4273fbc373592c166ec3.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen3-4B-Instruct-2507","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### HL-BT

Accuracy: 93.3. Result ID: 6a1d4273fbc373592c166ebf.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen3-4B-Instruct-2507","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### HL-BT

Accuracy: 93.2. Result ID: 6a1d4273fbc373592c166eba.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen2.5-3B-Instruct","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### Continual Offline Training RM

Accuracy: 93.1. Result ID: 6a1d4273fbc373592c166eb9.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Training Data":"HuggingFaceH4/ultrafeedback\_binarized"}. This record does not assert independent reproduction.

### Skywork-Reward-V2-Llama-3.2-3B

Accuracy: 93. Result ID: 6a1d4273fbc373592c166eb8.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79c61e209131cea7f4fe.md>)

Recorded evaluation configuration: {"Description":"Initial reward model without online updating"}. This record does not assert independent reproduction.

### Mean Reward

Accuracy: 83.4. Result ID: 6a1d4273fbc373592c166ec0.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen3-4B-Instruct-2507","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

### Mean Reward

Accuracy: 83.1. Result ID: 6a1d4273fbc373592c166ebb.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11dfdbd7150506f83dc6.md>)

Recorded evaluation configuration: {"Policy Model":"Qwen2.5-3B-Instruct","Response Generation Method":"GRPO"}. This record does not assert independent reproduction.

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