# Edit Inspectors Question Answering on MagicBrush (test) Source: [Edit Inspectors Question Answering on MagicBrush (test)]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:27:00.101Z Evaluation of model performance on the MagicBrush dataset (test split) for assessing the quality of image edits based on inspector questions. 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 | Edit Inspectors Question Answering | | Dataset (including any stated split) | MagicBrush (test) | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 7 of 7 | | 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](): Higher is better; unit: PERCENT. - [Contextual Consistency](): Higher is better; unit: PERCENT. - [Technical Precision](): Higher is better; unit: PERCENT. - [Artifacts Score](): Higher is better; unit: PERCENT. - [Diff Caption Acc](): Higher is better; unit: PERCENT. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | GPT-4 | 73.7 | Not recorded | 68fa79651e209131cea7d6a6 | | GPT-4o | 71 | Not recorded | 68fa79651e209131cea7d6a6 | | Qwen2.5 VL | 64 | Not recorded | 68fa79651e209131cea7d6a6 | | InternVL3 | 63.6 | Not recorded | 68fa79651e209131cea7d6a6 | | LLaVA | 62.3 | Not recorded | 68fa79651e209131cea7d6a6 | | GPT-4 Turbo | 60 | Not recorded | 68fa79651e209131cea7d6a6 | | LLaVA | 58.5 | Not recorded | 68fa79651e209131cea7d6a6 | ### GPT-4 Accuracy: 73.7. Result ID: 698b4b8e8d67a41d0a6ce145. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### GPT-4o Accuracy: 71. Result ID: 698b4b8e8d67a41d0a6ce146. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Qwen2.5 VL Accuracy: 64. Result ID: 698b4b8e8d67a41d0a6ce148. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### InternVL3 Accuracy: 63.6. Result ID: 698b4b8e8d67a41d0a6ce149. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### LLaVA Accuracy: 62.3. Result ID: 698b4b8e8d67a41d0a6ce14b. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"training":"Supervised"}. This record does not assert independent reproduction. ### GPT-4 Turbo Accuracy: 60. Result ID: 698b4b8e8d67a41d0a6ce147. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### LLaVA Accuracy: 58.5. Result ID: 698b4b8e8d67a41d0a6ce14a. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.