Benchmark Almanac

PRISM-Bench (image generation)

GPT-Image-1 86.3 leads since September 2025
Not in model cards

Models generate images from English or Chinese prompts. Vision-language judges score visual appeal and prompt alignment across seven tracks, including composition, lettering, and style.

The authors also released FLUX-Reason-6M for training. The evaluation prompts and code are in the benchmark repository.

Released
September 2025
Kind
Image generation and editing
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English, Chinese
In and out
Text → Images
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Last model added
September 2025
Scores refreshed
November 20th, 2026
18
Entries on its own leaderboard, 8 models we track

GPT-Image-1 leads with 86.3 among models released by September 2025 GPT-Image-1 leads with 86.3 among models released by September 2025

Best score per entry from the benchmark's own leaderboard, read on November 20th, 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. This view shows only models we track, each at its best.

The newest model with a score here came out in September 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-Image-1 GPT4.1, High 86.3
  2. 2 Gemini 2.5 Flash Image Preview (nano-banana) GPT4.1 85.3
  3. 3 Qwen-Image GPT4.1 79.9
  4. 4 Seedream 3.0 GPT4.1 79.6
  5. 5 HiDream-I1-Full GPT4.1 75.9
  6. 6 FLUX.1-Krea-dev GPT4.1 74.7
  7. 7 FLUX.1 [dev] GPT4.1 73.7
  8. 8 Stable Diffusion 3.5 Large GPT4.1 73.7
  9. 9 HiDream-I1-Dev GPT4.1 70.2
  10. 10 SD3.5-Medium GPT4.1 69.5
  11. 11 SD3-Medium Qwen2.5-VL 66.1
  12. 12 BAGEL GPT4.1, CoT 65.2
  13. 13 FLUX.1-schnell Qwen2.5-VL 64.7
  14. 14 Playground GPT4.1 64.1
  15. 15 Janus-Pro-7B Qwen2.5-VL 62.1
  16. 16 SDXL GPT4.1 60.4
  17. 17 SD2.1 Qwen2.5-VL 50.8
  18. 18 SD1.5 Qwen2.5-VL 46.0

Source: Official PRISM-Bench (image generation) leaderboard

#EntryScoreSettingReleasedRun onLinkSource
1 GPT-Image-1 86.3 GPT4.1, High Sep 2025 Sep 2025 platform.openai.com Official leaderboard
2 Gemini 2.5 Flash Image Preview (nano-banana) 85.3 GPT4.1 Sep 2025 Sep 2025 deepmind.google Official leaderboard
3 Qwen-Image 79.9 GPT4.1 Sep 2025 Sep 2025 huggingface.co Official leaderboard
4 Seedream 3.0 79.6 GPT4.1 Sep 2025 Sep 2025 seed.bytedance.com Official leaderboard
5 HiDream-I1-Full 75.9 GPT4.1 Sep 2025 huggingface.co Official leaderboard
6 FLUX.1-Krea-dev 74.7 GPT4.1 Sep 2025 huggingface.co Official leaderboard
7 FLUX.1 [dev] 73.7 GPT4.1 Sep 2025 Sep 2025 huggingface.co Official leaderboard
8 Stable Diffusion 3.5 Large 73.7 GPT4.1 Sep 2025 Sep 2025 huggingface.co Official leaderboard
9 HiDream-I1-Dev 70.2 GPT4.1 Sep 2025 huggingface.co Official leaderboard
10 SD3.5-Medium 69.5 GPT4.1 Sep 2025 huggingface.co Official leaderboard
11 SD3-Medium 66.1 Qwen2.5-VL Sep 2025 huggingface.co Official leaderboard
12 BAGEL 65.2 GPT4.1, CoT Sep 2025 Sep 2025 github.com Official leaderboard
13 FLUX.1-schnell 64.7 Qwen2.5-VL Sep 2025 huggingface.co Official leaderboard
14 Playground 64.1 GPT4.1 Sep 2025 huggingface.co Official leaderboard
15 Janus-Pro-7B 62.1 Qwen2.5-VL Jan 2025 Sep 2025 huggingface.co Official leaderboard
16 SDXL 60.4 GPT4.1 Sep 2025 huggingface.co Official leaderboard
17 SD2.1 50.8 Qwen2.5-VL Sep 2025 huggingface.co Official leaderboard
18 SD1.5 46.0 Qwen2.5-VL Sep 2025 huggingface.co Official leaderboard
#ModelScoreSettingReleasedRun onLinkSource
1 GPT-Image-1 86.3 GPT4.1, High Sep 2025 Sep 2025 platform.openai.com Official leaderboard
2 Gemini 2.5 Flash Image Preview (nano-banana) 85.3 GPT4.1 Sep 2025 Sep 2025 deepmind.google Official leaderboard
3 Qwen-Image 79.9 GPT4.1 Sep 2025 Sep 2025 huggingface.co Official leaderboard
4 Seedream 3.0 79.6 GPT4.1 Sep 2025 Sep 2025 seed.bytedance.com Official leaderboard
5 FLUX.1 [dev] 73.7 GPT4.1 Sep 2025 Sep 2025 huggingface.co Official leaderboard
6 Stable Diffusion 3.5 Large 73.7 GPT4.1 Sep 2025 Sep 2025 huggingface.co Official leaderboard
7 BAGEL 65.2 GPT4.1, CoT Sep 2025 Sep 2025 github.com Official leaderboard
8 Janus-Pro-7B 62.1 Qwen2.5-VL Jan 2025 Sep 2025 huggingface.co Official leaderboard

Rankings by part

We read 47 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Overall · Align, Overall · Aes and Imagination · Align. GPT-Image-1 leads 36 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.

Overall · Align

8 models we track have a score here.

  1. 1 Gemini 2.5 Flash Image Preview (nano-banana) ↑1 87.1
  2. 2 GPT-Image-1 ↓1 86.9
  3. 3 Qwen-Image 81.1
  4. 4 Seedream 3.0 80.5
  5. 5 Stable Diffusion 3.5 Large ↑1 73.9
  6. 6 FLUX.1 [dev] ↓1 72.4
  7. 7 BAGEL 67.5
  8. 8 Janus-Pro-7B 64.9

More effort adds a median 6 score

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
Qwen2.5-VL,… GPT4.1, High
86.3
Gemini 2.5 Flash Image Preview (nano-banana)
Qwen2.5-VL GPT4.1
85.3
Qwen-Image
Qwen2.5-VL GPT4.1
79.9
Seedream 3.0
Qwen2.5-VL GPT4.1
79.6
FLUX.1 [dev]
Qwen2.5-VL GPT4.1
73.7
Stable Diffusion 3.5 Large
Qwen2.5-VL GPT4.1
73.7
BAGEL
Qwen2.5-VL,… GPT4.1
65.2
Janus-Pro-7B
GPT4.1
62.1

Source: Official PRISM-Bench (image generation) leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-Image-1 GPT4.1, High 86.3
Gemini 2.5 Flash Image Preview (nano-banana) GPT4.1 85.3
GPT-Image-1 Qwen2.5-VL, High 80.7
Gemini 2.5 Flash Image Preview (nano-banana) Qwen2.5-VL 80.4
Qwen-Image GPT4.1 79.9
Seedream 3.0 GPT4.1 79.6
Seedream 3.0 Qwen2.5-VL 76.2
Qwen-Image Qwen2.5-VL 74.1
FLUX.1 [dev] GPT4.1 73.7
Stable Diffusion 3.5 Large GPT4.1 73.7
Stable Diffusion 3.5 Large Qwen2.5-VL 70.6
FLUX.1 [dev] Qwen2.5-VL 68.5
BAGEL GPT4.1, CoT 65.2
BAGEL GPT4.1 65.1
Janus-Pro-7B Qwen2.5-VL 62.1
BAGEL Qwen2.5-VL, CoT 62.0
BAGEL Qwen2.5-VL 62.0
Janus-Pro-7B GPT4.1 60.7

More effort adds a median 6 score

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
Qwen2.5-VL,… GPT4.1, High
86.3
Gemini 2.5 Flash Image Preview (nano-banana)
Qwen2.5-VL GPT4.1
85.3
Qwen-Image
Qwen2.5-VL GPT4.1
79.9
Seedream 3.0
Qwen2.5-VL GPT4.1
79.6
FLUX.1 [dev]
Qwen2.5-VL GPT4.1
73.7
Stable Diffusion 3.5 Large
Qwen2.5-VL GPT4.1
73.7
BAGEL
Qwen2.5-VL,… GPT4.1
65.2
Janus-Pro-7B
GPT4.1
62.1

Source: Official PRISM-Bench (image generation) leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-Image-1 GPT4.1, High 86.3
Gemini 2.5 Flash Image Preview (nano-banana) GPT4.1 85.3
GPT-Image-1 Qwen2.5-VL, High 80.7
Gemini 2.5 Flash Image Preview (nano-banana) Qwen2.5-VL 80.4
Qwen-Image GPT4.1 79.9
Seedream 3.0 GPT4.1 79.6
Seedream 3.0 Qwen2.5-VL 76.2
Qwen-Image Qwen2.5-VL 74.1
FLUX.1 [dev] GPT4.1 73.7
Stable Diffusion 3.5 Large GPT4.1 73.7
Stable Diffusion 3.5 Large Qwen2.5-VL 70.6
FLUX.1 [dev] Qwen2.5-VL 68.5
BAGEL GPT4.1, CoT 65.2
BAGEL GPT4.1 65.1
Janus-Pro-7B Qwen2.5-VL 62.1
BAGEL Qwen2.5-VL, CoT 62.0
BAGEL Qwen2.5-VL 62.0
Janus-Pro-7B GPT4.1 60.7

2 models have held number one on the official leaderboard since its release in September 2025

Lab claims count from the day the card came out, independent runs from the model’s release.

DeepSeek OpenAI
Own leaderboard Not tested since Sep 2025

Source: Official PRISM-Bench (image generation) leaderboard

Scores over time

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

Plot with other benchmarks
On the benchmark’s own leaderboard
60.0 70.0 80.0 90.0 2025 60.0 70.0 80.0 90.0 2025

Source: Official PRISM-Bench (image generation) leaderboard

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

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 14 months ago, in September 2025, for BAGEL. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (18 entries).
  • Pass Models scored 8 models we track, plus 10 other entries on its leaderboard.
  • Warning Newest score The newest came 14 months ago, in September 2025, for BAGEL.
  • 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 flux-reason-6m.github.io: 18 rows for models we track and 10 other entries at the last read.
  • Pass Last read November 20th, 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