# Few-Shot Science Question Answering on ARC Challenge

Source: [Few-Shot Science Question Answering on ARC Challenge](<https://www.sota2.com/research/sota/few-shot-science-question-answering-on-arc-challenge>)

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

Data updated: 2026-02-26T05:33:58.622Z

Evaluation of model performance on the ARC Challenge dataset using specific few-shot settings, measured by 0-shot and 32-shot accuracy.

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 | Science Question Answering |
| Dataset (including any stated split) | ARC Challenge |
| Metric | Accuracy (0-shot) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 6 of 6 |
| 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 (0-shot)](<https://www.sota2.com/research/sota/few-shot-science-question-answering-on-arc-challenge.md?metric=Accuracy%20(0-shot)>): Higher is better; unit: PERCENT.

- [Accuracy (32-shot)](<https://www.sota2.com/research/sota/few-shot-science-question-answering-on-arc-challenge.md?metric=Accuracy%20(32-shot)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (0-shot) | Result date | Report paper |
| --- | --- | --- | --- |
| OPT-IML 175B | 50.9 | Not recorded | 68fa6e791e209131cea401cb |
| OPT-IML 30B | 45.5 | Not recorded | 68fa6e791e209131cea401cb |
| OPT 175B | 44.1 | Not recorded | 68fa6e791e209131cea401cb |
| OPT 30B | 43.1 | Not recorded | 68fa6e791e209131cea401cb |
| OPT-IML 1.3B | 34 | Not recorded | 68fa6e791e209131cea401cb |
| OPT 1.3B | 33.9 | Not recorded | 68fa6e791e209131cea401cb |

### OPT-IML 175B

Accuracy (0-shot): 50.9. Result ID: 699023f73a6c314849fe648f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

Recorded evaluation configuration: {"Scale":"175B"}. This record does not assert independent reproduction.

### OPT-IML 30B

Accuracy (0-shot): 45.5. Result ID: 699023f73a6c314849fe648d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

Recorded evaluation configuration: {"Scale":"30B"}. This record does not assert independent reproduction.

### OPT 175B

Accuracy (0-shot): 44.1. Result ID: 699023f73a6c314849fe648e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"Scale":"175B"}. This record does not assert independent reproduction.

### OPT 30B

Accuracy (0-shot): 43.1. Result ID: 699023f73a6c314849fe648c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"Scale":"30B"}. This record does not assert independent reproduction.

### OPT-IML 1.3B

Accuracy (0-shot): 34. Result ID: 699023f73a6c314849fe648b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

Recorded evaluation configuration: {"Scale":"1.3B"}. This record does not assert independent reproduction.

### OPT 1.3B

Accuracy (0-shot): 33.9. Result ID: 699023f73a6c314849fe648a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa6e791e209131cea401cb.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6cda1e209131cea36474.md>)

[Implementation](<https://github.com/facebookresearch/metaseq>)

Recorded evaluation configuration: {"Scale":"1.3B"}. This record does not assert independent reproduction.

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