Image modelNo status listed

Seedream 4.0

ByteDance

Released: September 2025
Data date: August 15, 2026

Category view

Position within the category

This overview uses only published data from matching cohorts. Missing values never change a rank.

This position gives equal weight to 4 fixed Gradually image tests. One archived first output contributes for each model and task.

Position: Rank 25 of 28
Index score: 29.9 / 100
Coverage: 4 / 4

Leaderboard

1MAI-Image-2.5 Flash76.6 / 100

2GPT Image 271 / 100

3MAI-Image-2.567.6 / 100

24FLUX.2 Klein 9B35.2 / 100

25Seedream 4.029.9 / 100

26Midjourney V8.125.3 / 100

28Stable Diffusion 3.5 Large2.5 / 100

Source: Gradually-Bildtest

Measurements

Comparable benchmark results

Each chart contains exactly one source, one measurement series, and one stored comparison cohort. Bars show the position. The measured value appears on the right.

Text to Image Arena

1,227 Elo · Rank 13 of 25

Sample: 4,603 · Data date: August 26, 2026

1.GPT Image 21,371 Elo

11.MAI-Image-2.5 Flash1,229 Elo

12.HiDream-O1-Image-1.51,228 Elo

13.Seedream 4.01,227 Elo

14.FLUX.2 Max1,226 Elo

15.FLUX.2 Flex1,224 Elo

25.FLUX.2 Klein 9B1,144 Elo

7 of 25 model versions shown in this chart. A higher value ranks first.

Source: Artificial AnalysisParticipants: 25

Image Editing Arena

1,185 Elo · Rank 15 of 19

Sample: 10,984 · Data date: August 26, 2026

1.Reve 2.11,263 Elo

13.FLUX.2 Max1,201 Elo

14.HiDream-O1-Image1,190 Elo

15.Seedream 4.01,185 Elo

16.FLUX.2 Pro1,170 Elo

17.FLUX.2 Klein 9B1,166 Elo

19.FLUX.2 Flex1,162 Elo

7 of 19 model versions shown in this chart. A higher value ranks first.

Source: Artificial AnalysisParticipants: 19

GenExam Mathematics, strict

2.6% · Rank 6 of 6

Sample: 151 · Data date: August 26, 2026

1.Nano Banana 256.3%

2.Nano Banana Pro55.6%

3.GPT Image 250.3%

4.GPT Image 1.526.5%

5.FLUX.2 Max6.6%

6.Seedream 4.02.6%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Physics, strict

3.5% · Rank 6 of 6

Sample: 113 · Data date: August 26, 2026

1.GPT Image 279.6%

2.Nano Banana Pro75.2%

3.Nano Banana 274.3%

4.GPT Image 1.546%

5.FLUX.2 Max8.8%

6.Seedream 4.03.5%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Chemistry, strict

5.9% · Rank 6 of 6

Sample: 118 · Data date: August 26, 2026

1.GPT Image 269.5%

2.Nano Banana Pro60.2%

3.Nano Banana 252.5%

4.GPT Image 1.539%

5.FLUX.2 Max6.8%

6.Seedream 4.05.9%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Biology, strict

18.6% · Rank 5 of 6

Sample: 156 · Data date: August 26, 2026

1.GPT Image 289.1%

2.Nano Banana Pro75.6%

3.Nano Banana 266%

4.GPT Image 1.556.4%

5.Seedream 4.018.6%

6.FLUX.2 Max11.6%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Geography, strict

10.6% · Rank 6 of 6

Sample: 66 · Data date: August 26, 2026

1.GPT Image 284.8%

2.Nano Banana Pro75.8%

3.Nano Banana 269.7%

4.GPT Image 1.560.6%

5.FLUX.2 Max15.2%

6.Seedream 4.010.6%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Computer science, strict

6.9% · Rank 6 of 6

Sample: 102 · Data date: August 26, 2026

1.GPT Image 273.5%

2.Nano Banana Pro65.7%

3.Nano Banana 256.9%

4.GPT Image 1.536.3%

5.FLUX.2 Max8.8%

6.Seedream 4.06.9%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Engineering, strict

11.7% · Rank 5 of 6

Sample: 111 · Data date: August 26, 2026

1.GPT Image 279.3%

2.Nano Banana Pro71.2%

3.Nano Banana 267.6%

4.GPT Image 1.544.1%

5.Seedream 4.011.7%

6.FLUX.2 Max10.8%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Economics, strict

5.2% · Rank 5 of 6

Sample: 77 · Data date: August 26, 2026

1.Nano Banana Pro88.3%

2.GPT Image 283.1%

3.Nano Banana 263.6%

4.GPT Image 1.542.9%

5.Seedream 4.05.2%

6.FLUX.2 Max2.6%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Music, strict

0% · Rank 6 of 6

Sample: 65 · Data date: August 26, 2026

1.GPT Image 264.6%

2.Nano Banana Pro61.5%

3.Nano Banana 250.8%

4.GPT Image 1.529.2%

5.FLUX.2 Max6.2%

6.Seedream 4.00%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam History, strict

7.3% · Rank 5 of 6

Sample: 41 · Data date: August 26, 2026

1.Nano Banana Pro97.6%

2.GPT Image 282.9%

2.Nano Banana 282.9%

4.GPT Image 1.551.2%

5.FLUX.2 Max7.3%

5.Seedream 4.07.3%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam overall, strict

7.2% · Rank 6 of 6

Sample: 1,000 · Data date: August 26, 2026

1.GPT Image 274.6%

2.Nano Banana Pro72.7%

3.Nano Banana 264.1%

4.GPT Image 1.543.2%

5.FLUX.2 Max8.5%

6.Seedream 4.07.2%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Mathematics, relaxed

39.8% · Rank 6 of 6

Sample: 151 · Data date: August 26, 2026

1.Nano Banana 287.8%

2.Nano Banana Pro86.3%

3.GPT Image 285.2%

4.GPT Image 1.565.8%

5.FLUX.2 Max49.1%

6.Seedream 4.039.8%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Physics, relaxed

49% · Rank 6 of 6

Sample: 113 · Data date: August 26, 2026

1.Nano Banana 295.7%

2.GPT Image 295.6%

3.Nano Banana Pro95.1%

4.GPT Image 1.585.4%

5.FLUX.2 Max63.2%

6.Seedream 4.049%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Chemistry, relaxed

46.1% · Rank 6 of 6

Sample: 118 · Data date: August 26, 2026

1.GPT Image 292%

2.Nano Banana 290%

3.Nano Banana Pro88.7%

4.GPT Image 1.578.1%

5.FLUX.2 Max54%

6.Seedream 4.046.1%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Biology, relaxed

71% · Rank 6 of 6

Sample: 156 · Data date: August 26, 2026

1.GPT Image 297.5%

2.Nano Banana Pro95.9%

3.Nano Banana 295.2%

4.GPT Image 1.591.9%

5.FLUX.2 Max74.5%

6.Seedream 4.071%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Geography, relaxed

65.1% · Rank 6 of 6

Sample: 66 · Data date: August 26, 2026

1.GPT Image 297.6%

2.Nano Banana Pro96.5%

3.Nano Banana 294.8%

4.GPT Image 1.592.5%

5.FLUX.2 Max76.3%

6.Seedream 4.065.1%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Computer science, relaxed

52.2% · Rank 6 of 6

Sample: 102 · Data date: August 26, 2026

1.GPT Image 293.3%

2.Nano Banana Pro91.7%

3.Nano Banana 288.8%

4.GPT Image 1.575.8%

5.FLUX.2 Max56.5%

6.Seedream 4.052.2%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Engineering, relaxed

60% · Rank 6 of 6

Sample: 111 · Data date: August 26, 2026

1.GPT Image 296.5%

2.Nano Banana 295.8%

3.Nano Banana Pro95.1%

4.GPT Image 1.586.4%

5.FLUX.2 Max68.9%

6.Seedream 4.060%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Economics, relaxed

56% · Rank 6 of 6

Sample: 77 · Data date: August 26, 2026

1.GPT Image 297.7%

2.Nano Banana Pro97.2%

3.Nano Banana 294.2%

4.GPT Image 1.585.5%

5.FLUX.2 Max61.5%

6.Seedream 4.056%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam Music, relaxed

34.5% · Rank 6 of 6

Sample: 65 · Data date: August 26, 2026

1.Nano Banana Pro91%

2.GPT Image 289.1%

3.Nano Banana 286.9%

4.GPT Image 1.570.8%

5.FLUX.2 Max47%

6.Seedream 4.034.5%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam History, relaxed

56.7% · Rank 6 of 6

Sample: 41 · Data date: August 26, 2026

1.Nano Banana Pro99.9%

2.Nano Banana 297.3%

3.GPT Image 297.1%

4.GPT Image 1.590.9%

5.FLUX.2 Max68%

6.Seedream 4.056.7%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GenExam overall, relaxed

53% · Rank 6 of 6

Sample: 1,000 · Data date: August 26, 2026

1.GPT Image 293.8%

2.Nano Banana Pro93.7%

3.Nano Banana 292.6%

4.GPT Image 1.582.3%

5.FLUX.2 Max61.9%

6.Seedream 4.053%

6 of 6 model versions shown in this chart. A higher value ranks first.

1,000 multidisciplinary drawing tasks with reference images and fine-grained scoring points. Strict and relaxed scoring remain separate.

Source: GenExamParticipants: 6

GRADE reasoning

32.4 points · Rank 7 of 7

Sample: 520 · Data date: August 26, 2026

1.GPT Image 282.2 points

2.Nano Banana Pro77.5 points

3.Nano Banana 272.6 points

4.GPT Image 1.554.5 points

5.FLUX.2 Max47.8 points

6.FLUX.2 Pro38.9 points

7.Seedream 4.032.4 points

7 of 7 model versions shown in this chart. A higher value ranks first.

520 scientific image-editing tasks across ten academic domains. GRADE scores reasoning, consistency, readability, and domain accuracy separately.

Source: GRADEParticipants: 7

GRADE consistency

53.2 points · Rank 7 of 7

Sample: 520 · Data date: August 26, 2026

1.GPT Image 294.4 points

2.Nano Banana Pro89.5 points

3.Nano Banana 286.4 points

4.GPT Image 1.582.3 points

5.FLUX.2 Max67.2 points

6.FLUX.2 Pro55.5 points

7.Seedream 4.053.2 points

7 of 7 model versions shown in this chart. A higher value ranks first.

520 scientific image-editing tasks across ten academic domains. GRADE scores reasoning, consistency, readability, and domain accuracy separately.

Source: GRADEParticipants: 7

GRADE readability

77 points · Rank 5 of 7

Sample: 520 · Data date: August 26, 2026

1.GPT Image 298.8 points

2.Nano Banana 295.9 points

3.Nano Banana Pro95.8 points

4.GPT Image 1.590.7 points

5.Seedream 4.077 points

6.FLUX.2 Pro70.3 points

7.FLUX.2 Max68.6 points

7 of 7 model versions shown in this chart. A higher value ranks first.

520 scientific image-editing tasks across ten academic domains. GRADE scores reasoning, consistency, readability, and domain accuracy separately.

Source: GRADEParticipants: 7

GRADE accuracy

3.1 points · Rank 7 of 7

Sample: 520 · Data date: August 26, 2026

1.GPT Image 256 points

2.Nano Banana Pro46.2 points

3.Nano Banana 239.6 points

4.GPT Image 1.516 points

5.FLUX.2 Max11.9 points

6.FLUX.2 Pro4.4 points

7.Seedream 4.03.1 points

7 of 7 model versions shown in this chart. A higher value ranks first.

520 scientific image-editing tasks across ten academic domains. GRADE scores reasoning, consistency, readability, and domain accuracy separately.

Source: GRADEParticipants: 7

GEBench Chinese, single-step

62.04 points · Rank 5 of 5

Data date: August 26, 2026

1.Nano Banana Pro84.5 points

2.GPT Image 1.583.79 points

3.FLUX.2 Pro68.83 points

4.Wan 2.6 Text to Image64.2 points

5.Seedream 4.062.04 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench Chinese, multi-step

48.64 points · Rank 5 of 5

Data date: August 26, 2026

1.Nano Banana Pro68.65 points

2.GPT Image 1.556.97 points

3.FLUX.2 Pro55.07 points

4.Wan 2.6 Text to Image50.11 points

5.Seedream 4.048.64 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench Chinese, fictional app

49.28 points · Rank 5 of 5

Data date: August 26, 2026

1.Nano Banana Pro65.75 points

2.GPT Image 1.560.11 points

3.FLUX.2 Pro58.13 points

4.Wan 2.6 Text to Image52.72 points

5.Seedream 4.049.28 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench Chinese, real app

50.93 points · Rank 4 of 5

Data date: August 26, 2026

1.Nano Banana Pro64.35 points

2.GPT Image 1.555.65 points

3.FLUX.2 Pro55.41 points

4.Seedream 4.050.93 points

5.Wan 2.6 Text to Image50.4 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench Chinese, grounding

53.53 points · Rank 3 of 5

Data date: August 26, 2026

1.Nano Banana Pro64.83 points

2.Wan 2.6 Text to Image59.58 points

3.Seedream 4.053.53 points

4.GPT Image 1.553.33 points

5.FLUX.2 Pro50.24 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench Chinese, overall

52.88 points · Rank 5 of 5

Data date: August 26, 2026

1.Nano Banana Pro69.62 points

2.GPT Image 1.563.22 points

3.FLUX.2 Pro57.54 points

4.Wan 2.6 Text to Image55.4 points

5.Seedream 4.052.88 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench English, single-step

53.28 points · Rank 5 of 5

Data date: August 26, 2026

1.Nano Banana Pro84.32 points

2.GPT Image 1.580.8 points

3.FLUX.2 Pro61 points

4.Wan 2.6 Text to Image60.17 points

5.Seedream 4.053.28 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench English, multi-step

37.57 points · Rank 5 of 5

Data date: August 26, 2026

1.Nano Banana Pro69.51 points

2.GPT Image 1.558.87 points

3.FLUX.2 Pro52.17 points

4.Wan 2.6 Text to Image44.36 points

5.Seedream 4.037.57 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench English, fictional app

47.92 points · Rank 4 of 5

Data date: August 26, 2026

1.GPT Image 1.563.68 points

2.FLUX.2 Pro49.92 points

3.Wan 2.6 Text to Image49.55 points

4.Seedream 4.047.92 points

5.Nano Banana Pro46.33 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench English, real app

49.36 points · Rank 2 of 5

Data date: August 26, 2026

1.GPT Image 1.558.93 points

2.Seedream 4.049.36 points

3.Nano Banana Pro47.2 points

4.FLUX.2 Pro47.16 points

5.Wan 2.6 Text to Image44.8 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench English, grounding

44.17 points · Rank 5 of 5

Data date: August 26, 2026

1.Nano Banana Pro58.64 points

2.Wan 2.6 Text to Image53.36 points

3.GPT Image 1.549.23 points

4.FLUX.2 Pro45.67 points

5.Seedream 4.044.17 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

GEBench English, overall

46.46 points · Rank 5 of 5

Data date: August 26, 2026

1.GPT Image 1.563.16 points

2.Nano Banana Pro61.2 points

3.FLUX.2 Pro51.18 points

4.Wan 2.6 Text to Image50.45 points

5.Seedream 4.046.46 points

5 of 5 model versions shown in this chart. A higher value ranks first.

A bilingual graphical-user-interface benchmark. Chinese and English subsets and the single-step, multi-step, fictional-app, real-app, and grounding categories remain separate.

Source: GEBenchParticipants: 5

Gradually sample test: Typography and layout

18.5 rank points · Rank 23 of 28

Sample: 1 · Data date: August 18, 2026

1.MAI-Image-2.593.8 rank points

20.Reve 2.135.8 rank points

22.GPT Image 1.533.3 rank points

23.Seedream 4.018.5 rank points

24.Wan 2.6 Text to Image14.8 rank points

25.FLUX.2 Klein 9B9.9 rank points

28.Stable Diffusion 3.5 Large0 rank points

7 of 28 model versions shown in this chart. A higher value ranks first.

One archived, unfiltered first output per model. All 28 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.

Source: Gradually-BildtestParticipants: 28