# Educational Feedback Generation on SEFL (test)

Source: [Educational Feedback Generation on SEFL (test)](<https://www.sota2.com/research/sota/educational-feedback-generation-on-sefl-test>)

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

Data updated: 2026-02-26T06:08:57.451Z

Evaluation of model performance on the test split of the SEFL dataset for the Educational Feedback Generation task, measured by pairwise win rates against specific LLMs (GPT-4o, Claude 3.5 Sonnet, DeepSeek V3).

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 | Educational Feedback Generation |
| Dataset (including any stated split) | SEFL (test) |
| Metric | Win Rate (vs GPT-4o) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 2 of 2 |
| 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

- [Win Rate (vs GPT-4o)](<https://www.sota2.com/research/sota/educational-feedback-generation-on-sefl-test.md?metric=Win%20Rate%20(vs%20GPT-4o)>): Higher is better; unit: PERCENT.

- [Win Rate (vs Claude 3.5 Sonnet)](<https://www.sota2.com/research/sota/educational-feedback-generation-on-sefl-test.md?metric=Win%20Rate%20(vs%20Claude%203.5%20Sonnet)>): Higher is better; unit: PERCENT.

- [Win Rate (vs DeepSeek V3)](<https://www.sota2.com/research/sota/educational-feedback-generation-on-sefl-test.md?metric=Win%20Rate%20(vs%20DeepSeek%20V3)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Win Rate (vs GPT-4o) | Result date | Report paper |
| --- | --- | --- | --- |
| SEFL | 57 | Not recorded | 68fa7a561e209131cea82956 |
| Book2Dial | 43 | Not recorded | 68fa7a561e209131cea82956 |

### SEFL

Win Rate (vs GPT-4o): 57. Result ID: 699eb9ce5698665ec488cae9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a561e209131cea82956.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a561e209131cea82956.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B","Evaluation Protocol":"LLM-as-a-Judge"}. This record does not assert independent reproduction.

### Book2Dial

Win Rate (vs GPT-4o): 43. Result ID: 699eb9ce5698665ec488cae8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7a561e209131cea82956.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72621e209131cea561ac.md>)

[Implementation](<https://github.com/eth-lre/book2dial>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B","Evaluation Protocol":"LLM-as-a-Judge","Data size":"5,300 conversation pairs"}. This record does not assert independent reproduction.

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