# Instruction Following Evaluation on AlpacaEval 2.0 (test)

Source: [Instruction Following Evaluation on AlpacaEval 2.0 (test)](<https://www.sota2.com/research/sota/instruction-following-evaluation-on-alpacaeval-2-0-test>)

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

Data updated: 2026-03-04T11:54:28.491Z

Evaluation of model performance on instruction following using the test split of the AlpacaEval 2.0 dataset, measured by LC% over π0 and SE.

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 | Instruction Following Evaluation |
| Dataset (including any stated split) | AlpacaEval 2.0 (test) |
| Metric | LC% over π0 |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 10 of 10 |
| 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

- [LC% over π0](<https://www.sota2.com/research/sota/instruction-following-evaluation-on-alpacaeval-2-0-test.md?metric=LC%25%20over%20%CF%800>): Higher is better; unit: PERCENT.

- [SE](<https://www.sota2.com/research/sota/instruction-following-evaluation-on-alpacaeval-2-0-test.md?metric=SE>): Lower is better; unit: SCALAR.

- [WR (%)](<https://www.sota2.com/research/sota/instruction-following-evaluation-on-alpacaeval-2-0-test.md?metric=WR%20(%25)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | LC% over π0 | Result date | Report paper |
| --- | --- | --- | --- |
| DAR | 54.17 | Not recorded | 698eaff3c97a0b3e3b5e6617 |
| RLOO | 52.25 | Not recorded | 698eaff3c97a0b3e3b5e6617 |
| GRPO | 50.5 | Not recorded | 698eaff3c97a0b3e3b5e6617 |
| Iter-SFT | 49.8 | Not recorded | 698eaff3c97a0b3e3b5e6617 |
| Ours | 38.89 | Not recorded | 68fa7bef1e209131cea8b6e1 |
| Full Dataset | 34.37 | Not recorded | 68fa7bef1e209131cea8b6e1 |
| Random | 33.48 | Not recorded | 68fa7bef1e209131cea8b6e1 |
| Ours | 31.43 | Not recorded | 68fa7bef1e209131cea8b6e1 |
| Full Dataset | 29.24 | Not recorded | 68fa7bef1e209131cea8b6e1 |
| Random | 27.88 | Not recorded | 68fa7bef1e209131cea8b6e1 |

### DAR

LC% over π0: 54.17. Result ID: 6998bfe45698665ec422fcda.

[Reporting paper](<https://www.sota2.com/research/paper/698eaff3c97a0b3e3b5e6617.md>)

[Method origin paper](<https://www.sota2.com/research/paper/698eaff3c97a0b3e3b5e6617.md>)

Recorded evaluation configuration: {"Response Length":"1963"}. This record does not assert independent reproduction.

### RLOO

LC% over π0: 52.25. Result ID: 6998bfe45698665ec422fcd7.

[Reporting paper](<https://www.sota2.com/research/paper/698eaff3c97a0b3e3b5e6617.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa725a1e209131cea55df7.md>)

Recorded evaluation configuration: {"Response Length":"2076"}. This record does not assert independent reproduction.

### GRPO

LC% over π0: 50.5. Result ID: 6998bfe45698665ec422fcd8.

[Reporting paper](<https://www.sota2.com/research/paper/698eaff3c97a0b3e3b5e6617.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Response Length":"2038"}. This record does not assert independent reproduction.

### Iter-SFT

LC% over π0: 49.8. Result ID: 6998bfe45698665ec422fcd9.

[Reporting paper](<https://www.sota2.com/research/paper/698eaff3c97a0b3e3b5e6617.md>)

Recorded evaluation configuration: {"Response Length":"2004"}. This record does not assert independent reproduction.

### Ours

LC% over π0: 38.89. Result ID: 69a81d636dbb7075b94e143f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bef1e209131cea8b6e1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bef1e209131cea8b6e1.md>)

Recorded evaluation configuration: {"Base Model":"Llama-3.1-Tulu-3-8B","Data Selection":"30% easiest prompts"}. This record does not assert independent reproduction.

### Full Dataset

LC% over π0: 34.37. Result ID: 69a81d636dbb7075b94e143d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bef1e209131cea8b6e1.md>)

Recorded evaluation configuration: {"Base Model":"Llama-3.1-Tulu-3-8B","Data Selection":"Full Dataset"}. This record does not assert independent reproduction.

### Random

LC% over π0: 33.48. Result ID: 69a81d636dbb7075b94e143e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bef1e209131cea8b6e1.md>)

Recorded evaluation configuration: {"Base Model":"Llama-3.1-Tulu-3-8B","Data Selection":"30% random prompts"}. This record does not assert independent reproduction.

### Ours

LC% over π0: 31.43. Result ID: 69a81d636dbb7075b94e1442.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bef1e209131cea8b6e1.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bef1e209131cea8b6e1.md>)

Recorded evaluation configuration: {"Base Model":"Mistral-7B-Instruct-v0.2","Data Selection":"30% easiest prompts"}. This record does not assert independent reproduction.

### Full Dataset

LC% over π0: 29.24. Result ID: 69a81d636dbb7075b94e1440.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bef1e209131cea8b6e1.md>)

Recorded evaluation configuration: {"Base Model":"Mistral-7B-Instruct-v0.2","Data Selection":"Full Dataset"}. This record does not assert independent reproduction.

### Random

LC% over π0: 27.88. Result ID: 69a81d636dbb7075b94e1441.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bef1e209131cea8b6e1.md>)

Recorded evaluation configuration: {"Base Model":"Mistral-7B-Instruct-v0.2","Data Selection":"30% random prompts"}. This record does not assert independent reproduction.

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