# Visual Question Answering on SV-QA HRBench-8K

Source: [Visual Question Answering on SV-QA HRBench-8K](<https://www.sota2.com/research/sota/visual-question-answering-on-sv-qa-hrbench-8k>)

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

Data updated: 2026-05-20T08:03:50.739Z

Evaluation of model performance on the SV-QA HRBench-8K dataset variant for Visual Question Answering, reporting accuracy breakdowns for Q1, Q2, and combined correctness.

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 | Visual Question Answering |
| Dataset (including any stated split) | SV-QA HRBench-8K |
| Metric | Accuracy (Q1) |
| 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 (Q1)](<https://www.sota2.com/research/sota/visual-question-answering-on-sv-qa-hrbench-8k.md?metric=Accuracy%20(Q1)>): Higher is better; unit: PERCENT.

- [Accuracy (Q2)](<https://www.sota2.com/research/sota/visual-question-answering-on-sv-qa-hrbench-8k.md?metric=Accuracy%20(Q2)>): Higher is better; unit: PERCENT.

- [Accuracy (Both Correct)](<https://www.sota2.com/research/sota/visual-question-answering-on-sv-qa-hrbench-8k.md?metric=Accuracy%20(Both%20Correct)>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy (Q1) | Result date | Report paper |
| --- | --- | --- | --- |
| DeepEyes | 65.1 | Not recorded | 6a0d4002f984739ed6737774 |
| LVR | 62.9 | Not recorded | 6a0d4002f984739ed6737774 |
| SLVR-7B | 62.5 | Not recorded | 6a0d4002f984739ed6737774 |
| Visual-SR1 | 56.3 | Not recorded | 6a0d4002f984739ed6737774 |
| Qwen2.5-VL-7B | 53.5 | Not recorded | 6a0d4002f984739ed6737774 |
| GRIT | 49.1 | Not recorded | 6a0d4002f984739ed6737774 |

### DeepEyes

Accuracy (Q1): 65.1. Result ID: 6a0d6a64fbc373592c054c35.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78c71e209131cea7a75e.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Text + Visual Cropping"}. This record does not assert independent reproduction.

### LVR

Accuracy (Q1): 62.9. Result ID: 6a0d6a64fbc373592c054c37.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7be71e209131cea8b35b.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Visual Latent"}. This record does not assert independent reproduction.

### SLVR-7B

Accuracy (Q1): 62.5. Result ID: 6a0d6a64fbc373592c054c38.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Visual + Semantic Latent"}. This record does not assert independent reproduction.

### Visual-SR1

Accuracy (Q1): 56.3. Result ID: 6a0d6a64fbc373592c054c36.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ae11e209131cea85a1c.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Text CoT"}. This record does not assert independent reproduction.

### Qwen2.5-VL-7B

Accuracy (Q1): 53.5. Result ID: 6a0d6a64fbc373592c054c33.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Text-only"}. This record does not assert independent reproduction.

### GRIT

Accuracy (Q1): 49.1. Result ID: 6a0d6a64fbc373592c054c34.

[Reporting paper](<https://www.sota2.com/research/paper/6a0d4002f984739ed6737774.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78bd1e209131cea7a2a3.md>)

Recorded evaluation configuration: {"Reasoning Paradigm":"Text CoT"}. This record does not assert independent reproduction.

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