UniREditBench

Unified reasoning-based image editing benchmark covering real-world edits and game-world reasoning tasks.

14rows
overallprimary metric
2026-05-06sampled

Metadata

Metrics

Overall, Real-World-Overall, Viewpoint Transformation, Material Modification, Pose Adjustment, Temporal Evolution, Structural Integrity Change, Motion State Change, Spatial Arrangement, Mechanical Reaction, Medium Interaction, Game-World-Overall, 3D Reconstruction, Space Invader, Jewel2, Pacman, Word Search, Tictactoe, Sudoku, Maze, Sokoban

Latest Results

Rows are parsed from the public UniREditBench leaderboard JSON. Source model display names, links, Hugging Face URLs, and release metadata are preserved.

Rank Subject Overall Model Match Provenance Sampled
1 UniREdit-Bagel(Ours) 78.15 Imported 2026-05-06
2 GPT-4o 73.39 GPT-4o
openai-gpt-4o
Imported 2026-05-06
3 Nano Banana 68.26 Nano Banana (Gemini 2.5 Flash Image)
google-gemini-2.5-flash-image
Imported 2026-05-06
4 Wan2.5 61.36 Imported 2026-05-06
5 Qwen-Image-Edit 56.52 Imported 2026-05-06
6 Seedream4.0 55.77 Imported 2026-05-06
7 UniWorld-V2 54.87 Imported 2026-05-06
8 DreamOmni2 52.81 Imported 2026-05-06
9 Bagel-Think 50.96 Imported 2026-05-06
10 Step1X-Edit 50.15 Imported 2026-05-06
11 Lumina-DiMOO 48.54 Imported 2026-05-06
12 FLUX-Kontext-Pro 45.77 Imported 2026-05-06
13 OmniGen2 43.41 Imported 2026-05-06
14 MagicBrush 40.77 Imported 2026-05-06