# Negation Understanding on SimpleNeg COCO 2014 1.0 (val)

Source: [Negation Understanding on SimpleNeg COCO 2014 1.0 (val)](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val>)

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

Data updated: 2026-03-24T14:15:09.912Z

Evaluation of Negation Understanding performance on the validation split of the SimpleNeg COCO 2014 1.0 dataset, reporting various Top-K and Average metrics for Qwen models.

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 | Negation Understanding |
| Dataset (including any stated split) | SimpleNeg COCO 2014 1.0 (val) |
| Metric | Qwen3-VL-32B Top-1 Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

- [Qwen3-VL-32B Top-1 Accuracy](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-32B%20Top-1%20Accuracy>): Higher is better; unit: PERCENT.

- [Qwen3-VL-32B Average Score (Top 5)](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-32B%20Average%20Score%20(Top%205)>): Higher is better; unit: PERCENT.

- [Qwen3-VL-32B Top-5 Accuracy](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-32B%20Top-5%20Accuracy>): Higher is better; unit: PERCENT.

- [Qwen3-VL-8B Top-1 Accuracy](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-8B%20Top-1%20Accuracy>): Higher is better; unit: PERCENT.

- [Qwen3-VL-8B Average Score (Top 5)](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-8B%20Average%20Score%20(Top%205)>): Higher is better; unit: PERCENT.

- [Qwen3-VL-8B Top-5 Accuracy](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-8B%20Top-5%20Accuracy>): Higher is better; unit: PERCENT.

- [Qwen3-VL-4B Top-1 Accuracy](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-4B%20Top-1%20Accuracy>): Higher is better; unit: PERCENT.

- [Qwen3-VL-4B Average Score (Top 5)](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-4B%20Average%20Score%20(Top%205)>): Higher is better; unit: PERCENT.

- [Qwen3-VL-4B Top-5 Accuracy](<https://www.sota2.com/research/sota/negation-understanding-on-simpleneg-coco-2014-1-0-val.md?metric=Qwen3-VL-4B%20Top-5%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Qwen3-VL-32B Top-1 Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Steering | 54.3 | Not recorded | 69c21a5c0636e5bf4afc45ce |
| CLIP-CC12M | 53.1 | Not recorded | 69c21a5c0636e5bf4afc45ce |
| NegCLIP | 46 | Not recorded | 69c21a5c0636e5bf4afc45ce |
| Baseline | 44.9 | Not recorded | 69c21a5c0636e5bf4afc45ce |
| ConCLIP | 30.7 | Not recorded | 69c21a5c0636e5bf4afc45ce |

### Steering

Qwen3-VL-32B Top-1 Accuracy: 54.3. Result ID: 69c29af347f34ce924ab270d.

[Reporting paper](<https://www.sota2.com/research/paper/69c21a5c0636e5bf4afc45ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69c21a5c0636e5bf4afc45ce.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","# Negation Training Data":"4K","Without Finetuning?":"true"}. This record does not assert independent reproduction.

### CLIP-CC12M

Qwen3-VL-32B Top-1 Accuracy: 53.1. Result ID: 69c29af347f34ce924ab270c.

[Reporting paper](<https://www.sota2.com/research/paper/69c21a5c0636e5bf4afc45ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78841e209131cea78cb8.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","# Negation Training Data":"12M","Without Finetuning?":"false"}. This record does not assert independent reproduction.

### NegCLIP

Qwen3-VL-32B Top-1 Accuracy: 46. Result ID: 69c29af347f34ce924ab270b.

[Reporting paper](<https://www.sota2.com/research/paper/69c21a5c0636e5bf4afc45ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ee31e209131cea42a7d.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","# Negation Training Data":"120K","Without Finetuning?":"false"}. This record does not assert independent reproduction.

### Baseline

Qwen3-VL-32B Top-1 Accuracy: 44.9. Result ID: 69c29af347f34ce924ab2709.

[Reporting paper](<https://www.sota2.com/research/paper/69c21a5c0636e5bf4afc45ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69581e209131cea21ad0.md>)

[Implementation](<https://github.com/OpenAI/CLIP>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32"}. This record does not assert independent reproduction.

### ConCLIP

Qwen3-VL-32B Top-1 Accuracy: 30.7. Result ID: 69c29af347f34ce924ab270a.

[Reporting paper](<https://www.sota2.com/research/paper/69c21a5c0636e5bf4afc45ce.md>)

Recorded evaluation configuration: {"Backbone":"ViT-B/32","# Negation Training Data":"228K","Without Finetuning?":"false"}. This record does not assert independent reproduction.

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