TextSculpt-Bench - Text Accuracy: leaderboard
Text accuracy (0-100) of image editing models, from the full-image word edit distance between the expected and the observed text on TextSculpt-Bench (2026), 800 scene-text editing tasks (adding, removing, replacing and mixed edits of text in photos and posters). Higher is better.
Metric: Text accuracy (0-100): one minus the full-image word edit distance between the text GPT-5.2 infers should appear after editing and the text actually observed, normalized by the edited words and clipped at 0, over 800 TextSculpt-Bench text-editing tasks (200 each of addition, removal, replacement and hybrid editing) on Pexels photos and generated posters, default model settings; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 9 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seedream 4.5 | 76 | |
| 2 | Gemini 2.5 Flash Image | 63 | |
| 3 | FireRed-Image-Edit-1.0 | 62 | |
| 4 | Seedream 4.0 | 56 | |
| 5 | Qwen-Image-Edit-2511 | 55 | |
| 6 | LongCat-Image-Edit | 54 | |
| 7 | Step1X-Edit | 47 | |
| 8 | Bagel | 26 | |
| 9 | OmniGen2 | 21 | #4625 |
Interactive version: theaggregate.ai/benchmark?slug=textsculpt-bench-text-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-22.