# Multimodal Question Answering on ScienceQA (SQA) (test)

Source: [Multimodal Question Answering on ScienceQA (SQA) (test)](<https://www.sota2.com/research/sota/multimodal-question-answering-on-scienceqa-sqa-test>)

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

Data updated: 2026-02-26T06:03:03.136Z

Evaluation of model performance on the test split of the ScienceQA (SQA) variant dataset for Multimodal Question Answering, reporting SQA 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 | Multimodal Question Answering |
| Dataset (including any stated split) | ScienceQA (SQA) (test) |
| Metric | SQA Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 4 of 4 |
| 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

- [SQA Accuracy](<https://www.sota2.com/research/sota/multimodal-question-answering-on-scienceqa-sqa-test.md?metric=SQA%20Accuracy>): Higher is better; unit: PERCENT.

- [CUDA Time](<https://www.sota2.com/research/sota/multimodal-question-answering-on-scienceqa-sqa-test.md?metric=CUDA%20Time>): Lower is better; unit: TIME.

- [CUDA Time Reduction](<https://www.sota2.com/research/sota/multimodal-question-answering-on-scienceqa-sqa-test.md?metric=CUDA%20Time%20Reduction>): Higher is better; unit: PERCENT.

- [FLOPs (T)](<https://www.sota2.com/research/sota/multimodal-question-answering-on-scienceqa-sqa-test.md?metric=FLOPs%20(T)>): Lower is better; unit: SCALAR.

- [FLOPs Reduction](<https://www.sota2.com/research/sota/multimodal-question-answering-on-scienceqa-sqa-test.md?metric=FLOPs%20Reduction>): Higher is better; unit: PERCENT.

- [KV Cache (MB)](<https://www.sota2.com/research/sota/multimodal-question-answering-on-scienceqa-sqa-test.md?metric=KV%20Cache%20(MB)>): Lower is better; unit: SIZE.

- [KV Cache Reduction](<https://www.sota2.com/research/sota/multimodal-question-answering-on-scienceqa-sqa-test.md?metric=KV%20Cache%20Reduction>): Higher is better; unit: PERCENT.

## Recorded results

| Method | SQA Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Vanilla (LLaVA-NeXT-7B) | 70.2 | Not recorded | 69802f6ba8b01e76cc8e14f5 |
| VisionTrim | 69.6 | Not recorded | 69802f6ba8b01e76cc8e14f5 |
| VisionZip | 67.3 | Not recorded | 69802f6ba8b01e76cc8e14f5 |
| SparseVLM | 67 | Not recorded | 69802f6ba8b01e76cc8e14f5 |

### Vanilla (LLaVA-NeXT-7B)

SQA Accuracy: 70.2. Result ID: 699773715698665ec4c77f6d.

[Reporting paper](<https://www.sota2.com/research/paper/69802f6ba8b01e76cc8e14f5.md>)

Recorded evaluation configuration: {"#Tokens":"2880"}. This record does not assert independent reproduction.

### VisionTrim

SQA Accuracy: 69.6. Result ID: 699773715698665ec4c77f70.

[Reporting paper](<https://www.sota2.com/research/paper/69802f6ba8b01e76cc8e14f5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69802f6ba8b01e76cc8e14f5.md>)

[Implementation](<https://github.com/hanxunyu/VisionTrim>)

Recorded evaluation configuration: {"#Tokens":"320"}. This record does not assert independent reproduction.

### VisionZip

SQA Accuracy: 67.3. Result ID: 699773715698665ec4c77f6f.

[Reporting paper](<https://www.sota2.com/research/paper/69802f6ba8b01e76cc8e14f5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"#Tokens":"320"}. This record does not assert independent reproduction.

### SparseVLM

SQA Accuracy: 67. Result ID: 699773715698665ec4c77f6e.

[Reporting paper](<https://www.sota2.com/research/paper/69802f6ba8b01e76cc8e14f5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79181e209131cea7c1b1.md>)

Recorded evaluation configuration: {"#Tokens":"320"}. This record does not assert independent reproduction.

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