Benchmark Almanac

GEBench

Nano Banana Pro 69.6 tops its leaderboard GPT-Image-1.5 63.2 leads the models we track
Not in model cards

Image generation models create interactive GUI states from prompts. A judge scores goal, logic, consistency, interface and quality across five task types.

The release contains 700 samples in five categories and reports Chinese and English results for 12 models.

Made by
StepFun and collaborators
Released
February 2026
Kind
General visual questions
Level
Everyday
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Measures
Capability
Language
Many languages
In and out
Text → Images
Interaction
Single turn
Items from
Generated
Refresh
One fixed set
Run by
Anyone
Scores refreshed
November 21st, 2026
12
Entries on its own leaderboard, 8 models we track

Nano Banana Pro tops the leaderboard with 69.6 GPT-Image-1.5 leads with 63.2

Best score per entry from the benchmark's own leaderboard, read on November 21st, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is GPT-Image-1.5, at 63.2. This view shows only models we track, each at its best.

  1. 1 Nano Banana Pro Chinese 69.6
  2. 2 GPT-Image-1.5 Chinese 63.2
  3. 3 FLUX.2 [pro] Chinese 57.5
  4. 4 Nano Banana Chinese 56.7
  5. 5 Seedream 4.5 Chinese 55.9
  6. 6 GPT-Image-1 English 55.6
  7. 7 Wan 2.6 Chinese 55.4
  8. 8 Seedream 4.0 Chinese 52.9
  9. 9 UniWorld-V2 Chinese 36.7
  10. 10 Longcat-Image Chinese 34.5
  11. 11 Qwen-Image-Edit Chinese 33.7
  12. 12 BAGEL Chinese 28.9

Source: Official GEBench leaderboard

#EntryScoreSettingDateSource
1 Nano Banana Pro 69.6 Chinese Official leaderboard
2 GPT-Image-1.5 63.2 Chinese Official leaderboard
3 FLUX.2 [pro] 57.5 Chinese Official leaderboard
4 Nano Banana 56.7 Chinese Official leaderboard
5 Seedream 4.5 55.9 Chinese Official leaderboard
6 GPT-Image-1 55.6 English Official leaderboard
7 Wan 2.6 55.4 Chinese Official leaderboard
8 Seedream 4.0 52.9 Chinese Official leaderboard
9 UniWorld-V2 36.7 Chinese Official leaderboard
10 Longcat-Image 34.5 Chinese Official leaderboard
11 Qwen-Image-Edit 33.7 Chinese Official leaderboard
12 BAGEL 28.9 Chinese Official leaderboard
#ModelScoreSettingDateSource
1 GPT-Image-1.5 63.2 Chinese Official leaderboard
2 FLUX.2 [pro] 57.5 Chinese Official leaderboard
3 Seedream 4.5 55.9 Chinese Official leaderboard
4 GPT-Image-1 55.6 English Official leaderboard
5 Wan 2.6 55.4 Chinese Official leaderboard
6 Seedream 4.0 52.9 Chinese Official leaderboard
7 Qwen-Image-Edit 33.7 Chinese Official leaderboard
8 BAGEL 28.9 Chinese Official leaderboard

Rankings by part

We read 5 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Single-Step, Multi-Step and Fiction-App. GPT-Image-1.5 leads 2 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Single-Step

8 models we track have a score here.

  1. 1 GPT-Image-1.5 83.8
  2. 2 FLUX.2 [pro] 68.8
  3. 3 GPT-Image-1 ↑1 64.7
  4. 4 Wan 2.6 ↑1 64.2
  5. 5 Seedream 4.5 ↓2 63.6
  6. 6 Seedream 4.0 62.0
  7. 7 Qwen-Image-Edit 41.1
  8. 8 BAGEL 34.8

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-Image-1.5
English
63.2
FLUX.2 [pro]
English Chinese
57.5
Seedream 4.5
English Chinese
55.9
GPT-Image-1
Chinese
55.6
Wan 2.6
English Chinese
55.4
Seedream 4.0
English Chinese
52.9
Qwen-Image-Edit
English
33.7
BAGEL
English
28.9

Source: Official GEBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-Image-1.5 Chinese 63.2
GPT-Image-1.5 English 63.2
FLUX.2 [pro] Chinese 57.5
Seedream 4.5 Chinese 55.9
GPT-Image-1 English 55.6
Wan 2.6 Chinese 55.4
Seedream 4.0 Chinese 52.9
GPT-Image-1 Chinese 52.4
FLUX.2 [pro] English 51.2
Wan 2.6 English 50.5
Seedream 4.5 English 50.0
Seedream 4.0 English 46.5
Qwen-Image-Edit Chinese 33.7
Qwen-Image-Edit English 33.0
BAGEL Chinese 28.9
BAGEL English 28.0

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-Image-1.5
English
63.2
FLUX.2 [pro]
English Chinese
57.5
Seedream 4.5
English Chinese
55.9
GPT-Image-1
Chinese
55.6
Wan 2.6
English Chinese
55.4
Seedream 4.0
English Chinese
52.9
Qwen-Image-Edit
English
33.7
BAGEL
English
28.9

Source: Official GEBench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-Image-1.5 Chinese 63.2
GPT-Image-1.5 English 63.2
FLUX.2 [pro] Chinese 57.5
Seedream 4.5 Chinese 55.9
GPT-Image-1 English 55.6
Wan 2.6 Chinese 55.4
Seedream 4.0 Chinese 52.9
GPT-Image-1 Chinese 52.4
FLUX.2 [pro] English 51.2
Wan 2.6 English 50.5
Seedream 4.5 English 50.0
Seedream 4.0 English 46.5
Qwen-Image-Edit Chinese 33.7
Qwen-Image-Edit English 33.0
BAGEL Chinese 28.9
BAGEL English 28.0

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

No dated values yet.

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Vision

All 636 in Vision
  • MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy. Qwen 3.6 Plus 86%
  • MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
  • CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
  • MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy. Seed2.1 Pro 90.7%
  • MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy. Qwen 3.8 Max 97.7%
  • MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy. Qwen 3.8 Max 85.9%

Data audit

All 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • Pass Independent runs Scores from the official leaderboard (12 entries).
  • Pass Models scored 8 models we track, plus 4 other entries on its leaderboard.
  • Note Newest score No scores yet.
  • Note Frontier models None of its models are on the capabilities index.
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on stepfun-ai.github.io: 16 rows for models we track and 4 other entries at the last read.
  • Pass Last read November 21st, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check