# Multi-judge evaluation on Shared 500-prompt sample

Source: [Multi-judge evaluation on Shared 500-prompt sample](<https://www.sota2.com/research/sota/multi-judge-evaluation-on-shared-500-prompt-sample>)

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

Data updated: 2026-03-16T08:50:16.085Z

Evaluation of model performance using a multi-judge setup on the Shared 500-prompt sample, focusing on correlation and gap metrics.

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 | Multi-judge evaluation |
| Dataset (including any stated split) | Shared 500-prompt sample |
| Metric | Global Correlation (r) |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

- [Global Correlation (r)](<https://www.sota2.com/research/sota/multi-judge-evaluation-on-shared-500-prompt-sample.md?metric=Global%20Correlation%20(r)>): Higher is better; unit: SCALAR.

- [Within-Judge Correlation (r)](<https://www.sota2.com/research/sota/multi-judge-evaluation-on-shared-500-prompt-sample.md?metric=Within-Judge%20Correlation%20(r)>): Higher is better; unit: SCALAR.

- [Agreement Gap (%)](<https://www.sota2.com/research/sota/multi-judge-evaluation-on-shared-500-prompt-sample.md?metric=Agreement%20Gap%20(%25)>): Lower is better; unit: PERCENT.

- [Recovery Rate (%)](<https://www.sota2.com/research/sota/multi-judge-evaluation-on-shared-500-prompt-sample.md?metric=Recovery%20Rate%20(%25)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Global Correlation (r) | Result date | Report paper |
| --- | --- | --- | --- |
| GPT-5.2 | 0.87 | Not recorded | 69b78e9e0636e5bf4afbdcc9 |
| Claude Sonnet 4 | 0.59 | Not recorded | 69b78e9e0636e5bf4afbdcc9 |
| GPT-4.1-mini | 0.56 | Not recorded | 69b78e9e0636e5bf4afbdcc9 |
| Gemini-2.5-flash | 0.47 | Not recorded | 69b78e9e0636e5bf4afbdcc9 |
| Llama-3.3-70b | 0.31 | Not recorded | 69b78e9e0636e5bf4afbdcc9 |

### GPT-5.2

Global Correlation (r): 0.87. Result ID: 69b7c43147f34ce9249f7f85.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e9e0636e5bf4afbdcc9.md>)

Recorded evaluation configuration: {"Family":"OpenAI"}. This record does not assert independent reproduction.

### Claude Sonnet 4

Global Correlation (r): 0.59. Result ID: 69b7c43147f34ce9249f7f86.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e9e0636e5bf4afbdcc9.md>)

Recorded evaluation configuration: {"Family":"Anthropic"}. This record does not assert independent reproduction.

### GPT-4.1-mini

Global Correlation (r): 0.56. Result ID: 69b7c43147f34ce9249f7f87.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e9e0636e5bf4afbdcc9.md>)

Recorded evaluation configuration: {"Family":"OpenAI"}. This record does not assert independent reproduction.

### Gemini-2.5-flash

Global Correlation (r): 0.47. Result ID: 69b7c43147f34ce9249f7f88.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e9e0636e5bf4afbdcc9.md>)

Recorded evaluation configuration: {"Family":"Google"}. This record does not assert independent reproduction.

### Llama-3.3-70b

Global Correlation (r): 0.31. Result ID: 69b7c43147f34ce9249f7f89.

[Reporting paper](<https://www.sota2.com/research/paper/69b78e9e0636e5bf4afbdcc9.md>)

Recorded evaluation configuration: {"Family":"Meta"}. This record does not assert independent reproduction.

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