{
  "schemaVersion": 1,
  "benchmark": {
    "id": "geditbench-2",
    "name": "GEditBench",
    "version": "2",
    "category": "image",
    "question": "Can it make an edit without breaking the rest?",
    "summary": "Instruction following, visual quality, and preservation of the original image across 23 editing tasks.",
    "source": {
      "name": "GEditBench v2 team",
      "url": "https://zhangqijiang07.github.io/gedit2_web/",
      "methodologyUrl": "https://github.com/ZhangqiJiang07/GEditBench_v2"
    },
    "measure": "Overall pairwise Elo; higher is better. Publisher bootstrap intervals included.",
    "comparisonRule": "GPT-4o judges instruction and quality; PVC-Judge scores consistency. Compare within this fixed evaluation pool.",
    "limitations": [
      "Automated, human-aligned judging is not direct human voting.",
      "This is the 2026 paper cohort; the dated API names are retained.",
      "Safety refusals and failures leave some models with fewer evaluated samples.",
      "Elo is pool-relative, not a percentage; overlapping intervals do not establish a winner."
    ],
    "coverage": "charted",
    "tags": [
      "image editing",
      "identity",
      "preservation",
      "instruction following",
      "FLUX",
      "Seedream",
      "Qwen Image Edit"
    ]
  },
  "dataset": {
    "benchmarkId": "geditbench-2",
    "version": "2",
    "score": {
      "label": "Editing overall",
      "unit": "Elo",
      "direction": "higher"
    },
    "source": {
      "name": "GEditBench v2 team",
      "url": "https://zhangqijiang07.github.io/gedit2_web/",
      "retrievedAt": "2026-09-17T15:07:07.972Z",
      "revision": "sha256:87a97fa869d8d879e2f834520a109632e41c7f8e43724aabae6bb13882f838cc"
    },
    "evidenceLabel": "2026 paper cohort · model-judged",
    "configurationLabel": "Image-editing model · fixed judge pool",
    "comparabilityNote": "GPT-4o judges instruction following and quality; PVC-Judge scores preservation. Elo and bootstrap intervals belong only to this 16-model research pool. Evaluated sample counts vary by model.",
    "points": [
      {
        "id": "nano-banana-pro-26-03-04",
        "label": "Nano Banana Pro (26-03-04)",
        "model": "Nano Banana Pro (26-03-04)",
        "provider": "Google",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1096,
        "costUsd": null,
        "uncertainty": {
          "lower": 1090,
          "upper": 1102,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1126"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1066"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1108"
          },
          {
            "label": "Evaluated samples",
            "value": "1156"
          }
        ]
      },
      {
        "id": "seedream-4-5-26-03-11",
        "label": "Seedream 4.5 (26-03-11)",
        "model": "Seedream 4.5 (26-03-11)",
        "provider": "ByteDance",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1089,
        "costUsd": null,
        "uncertainty": {
          "lower": 1082,
          "upper": 1096,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1111"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1142"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1030"
          },
          {
            "label": "Evaluated samples",
            "value": "1190"
          }
        ]
      },
      {
        "id": "gpt-image-1-5-26-03-04",
        "label": "GPT Image 1.5 (26-03-04)",
        "model": "GPT Image 1.5 (26-03-04)",
        "provider": "OpenAI",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1071,
        "costUsd": null,
        "uncertainty": {
          "lower": 1064,
          "upper": 1077,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1260"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1149"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "846"
          },
          {
            "label": "Evaluated samples",
            "value": "1081"
          }
        ]
      },
      {
        "id": "flux-2-klein-9b",
        "label": "FLUX.2 [klein] 9B",
        "model": "FLUX.2 [klein] 9B",
        "provider": "Black Forest Labs",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1039,
        "costUsd": null,
        "uncertainty": {
          "lower": 1033,
          "upper": 1045,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1083"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1025"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1019"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "qwen-image-edit-2511",
        "label": "Qwen-Image-Edit-2511",
        "model": "Qwen-Image-Edit-2511",
        "provider": "Alibaba",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1038,
        "costUsd": null,
        "uncertainty": {
          "lower": 1032,
          "upper": 1044,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1095"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1060"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "972"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "flux-2-klein-4b",
        "label": "FLUX.2 [klein] 4B",
        "model": "FLUX.2 [klein] 4B",
        "provider": "Black Forest Labs",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1031,
        "costUsd": null,
        "uncertainty": {
          "lower": 1025,
          "upper": 1037,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1007"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1019"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1070"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "flux-2-dev-turbo",
        "label": "FLUX.2 [dev] Turbo",
        "model": "FLUX.2 [dev] Turbo",
        "provider": "Black Forest Labs",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1021,
        "costUsd": null,
        "uncertainty": {
          "lower": 1015,
          "upper": 1027,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1068"
          },
          {
            "label": "Visual quality · Elo",
            "value": "936"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1064"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "qwen-image-edit-2509",
        "label": "Qwen-Image-Edit-2509",
        "model": "Qwen-Image-Edit-2509",
        "provider": "Alibaba",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1014,
        "costUsd": null,
        "uncertainty": {
          "lower": 1009,
          "upper": 1020,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1033"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1062"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "955"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "qwen-image-edit",
        "label": "Qwen-Image-Edit",
        "model": "Qwen-Image-Edit",
        "provider": "Alibaba",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1010,
        "costUsd": null,
        "uncertainty": {
          "lower": 1004,
          "upper": 1016,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "991"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1073"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "971"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "flux-2-dev",
        "label": "FLUX.2 [dev]",
        "model": "FLUX.2 [dev]",
        "provider": "Black Forest Labs",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1006,
        "costUsd": null,
        "uncertainty": {
          "lower": 999,
          "upper": 1013,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1037"
          },
          {
            "label": "Visual quality · Elo",
            "value": "965"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1018"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "longcat-image-edit",
        "label": "LongCat-Image-Edit",
        "model": "LongCat-Image-Edit",
        "provider": "Meituan",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 1001,
        "costUsd": null,
        "uncertainty": {
          "lower": 995,
          "upper": 1006,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "1018"
          },
          {
            "label": "Visual quality · Elo",
            "value": "968"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1017"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "step1x-edit-v1p2",
        "label": "Step1X-Edit-v1p2",
        "model": "Step1X-Edit-v1p2",
        "provider": "StepFun",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 996,
        "costUsd": null,
        "uncertainty": {
          "lower": 990,
          "upper": 1003,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "909"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1007"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1067"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "glm-image",
        "label": "GLM-Image",
        "model": "GLM-Image",
        "provider": "Z.ai",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 979,
        "costUsd": null,
        "uncertainty": {
          "lower": 973,
          "upper": 985,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "787"
          },
          {
            "label": "Visual quality · Elo",
            "value": "1023"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "1109"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "omnigen-v2",
        "label": "OmniGen V2",
        "model": "OmniGen V2",
        "provider": "OmniGen V2 project",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 888,
        "costUsd": null,
        "uncertainty": {
          "lower": 881,
          "upper": 895,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "807"
          },
          {
            "label": "Visual quality · Elo",
            "value": "910"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "929"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "flux-1-kontext-dev",
        "label": "FLUX.1 Kontext [dev]",
        "model": "FLUX.1 Kontext [dev]",
        "provider": "Black Forest Labs",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 869,
        "costUsd": null,
        "uncertainty": {
          "lower": 862,
          "upper": 877,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "849"
          },
          {
            "label": "Visual quality · Elo",
            "value": "900"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "840"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      },
      {
        "id": "bagel",
        "label": "Bagel",
        "model": "Bagel",
        "provider": "ByteDance",
        "harness": "GEditBench v2 · GPT-4o / PVC-Judge",
        "effort": null,
        "score": 851,
        "costUsd": null,
        "uncertainty": {
          "lower": 843,
          "upper": 859,
          "label": "Publisher confidence interval · 1,000 bootstrap iterations"
        },
        "sourceUrl": "https://zhangqijiang07.github.io/gedit2_web/",
        "details": [
          {
            "label": "Instruction following · Elo",
            "value": "820"
          },
          {
            "label": "Visual quality · Elo",
            "value": "694"
          },
          {
            "label": "Visual consistency · Elo",
            "value": "987"
          },
          {
            "label": "Evaluated samples",
            "value": "1200"
          }
        ]
      }
    ]
  }
}