Arena-T2I Hard
Models generate images from detailed user prompts. A judge checks whether the images satisfy individual requirements and their dependencies.
The authors evaluated 11 systems on 310 prompts.
- Made by
- Arena
- Released
- June 2026
- Kind
- Image generation and editing
- Level
- University
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Image, video and 3D generation
- Scores refreshed
- November 20th, 2026
Gemini 3 Pro Image (nano-banana-pro, 2K) leads with 0.85 Gemini 3 Pro Image (nano-banana-pro, 2K) leads with 0.85
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. 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.
- 0.85
Gemini 3 Pro Image (nano-banana-pro, 2K)
- 0.85 Grok Imagine Image
- 0.80 GPT-Image-1.5
- 0.79 Recraft V4
- 0.77
Gemini 2.5 Flash Image Preview (nano-banana)
- 0.77 wan2.6-t2i-v2
- 0.72 GPT-Image-1
- 0.68
Imagen 4 Ultra (generate-001)
- 0.66
Imagen 4 (generate-001)
- 0.61
HunyuanImage 3.0
- 0.53 FLUX.1 [dev]
- 0.52 Ideogram 3.0
- 0.41 SD3.5-M
- 1
Gemini 3 Pro Image (nano-banana-pro, 2K) 0.85
- 2 Grok Imagine Image 0.85
- 3 GPT-Image-1.5 High fidelity 0.80
- 4 Recraft V4 0.79
- 5
Gemini 2.5 Flash Image Preview (nano-banana) 0.77
- 6 wan2.6-t2i-v2 0.77
- 7 GPT-Image-1 0.72
- 8
Imagen 4 Ultra (generate-001) 0.68
- 9
Imagen 4 (generate-001) 0.66
- 10
HunyuanImage 3.0 0.61
- 11 FLUX.1 [dev] Faith + Pick (ours) 0.53
- 12 Ideogram 3.0 Quality 0.52
- 13 SD3.5-M Faith + Pick (ours) 0.41
Source: The Arena-T2I Hard paper
- 1
Gemini 3 Pro Image (nano-banana-pro, 2K) 0.85
- 2 Grok Imagine Image 0.85
- 3 GPT-Image-1.5 High fidelity 0.80
- 4 Recraft V4 0.79
- 5
Gemini 2.5 Flash Image Preview (nano-banana) 0.77
- 6 GPT-Image-1 0.72
- 7
Imagen 4 Ultra (generate-001) 0.68
- 8
Imagen 4 (generate-001) 0.66
- 9
HunyuanImage 3.0 0.61
- 10 FLUX.1 [dev] Faith + Pick (ours) 0.53
- 11 Ideogram 3.0 Quality 0.52
Source: The Arena-T2I Hard paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 0.85 | Official leaderboard | ||
| 2 | Grok Imagine Image | 0.85 | Official leaderboard | ||
| 3 | GPT-Image-1.5 | 0.80 | High fidelity | Official leaderboard | |
| 4 | Recraft V4 | 0.79 | Official leaderboard | ||
| 5 | | 0.77 | Official leaderboard | ||
| 6 | wan2.6-t2i-v2 | 0.77 | Official leaderboard | ||
| 7 | GPT-Image-1 | 0.72 | Official leaderboard | ||
| 8 | | 0.68 | Official leaderboard | ||
| 9 | | 0.66 | Official leaderboard | ||
| 10 | | 0.61 | Official leaderboard | ||
| 11 | FLUX.1 [dev] | 0.53 | Faith + Pick (ours) | Official leaderboard | |
| 12 | Ideogram 3.0 | 0.52 | Quality | Official leaderboard | |
| 13 | SD3.5-M | 0.41 | Faith + Pick (ours) | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 0.85 | Official leaderboard | ||
| 2 | Grok Imagine Image | 0.85 | Official leaderboard | ||
| 3 | GPT-Image-1.5 | 0.80 | High fidelity | Official leaderboard | |
| 4 | Recraft V4 | 0.79 | Official leaderboard | ||
| 5 | | 0.77 | Official leaderboard | ||
| 6 | GPT-Image-1 | 0.72 | Official leaderboard | ||
| 7 | | 0.68 | Official leaderboard | ||
| 8 | | 0.66 | Official leaderboard | ||
| 9 | | 0.61 | Official leaderboard | ||
| 10 | FLUX.1 [dev] | 0.53 | Faith + Pick (ours) | Official leaderboard | |
| 11 | Ideogram 3.0 | 0.52 | Quality | Official leaderboard |
Rankings by part
We read 9 parts from the benchmark’s paper, each ranking models on one part of the test, such as 3D modeling, Art and Cartoon. Gemini 3 Pro Image (nano-banana-pro, 2K) leads 5 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.
3D modeling
10 models we track have a score here.
- 1
Gemini 3 Pro Image (nano-banana-pro, 2K) 0.86
- 2 GPT-Image-1.5 ↑1 0.82
- 3 Grok Imagine Image ↓1 0.79
- 4
Gemini 2.5 Flash Image Preview (nano-banana) ↑1 0.76
- 5 Recraft V4 ↓1 0.72
- 6 GPT-Image-1 0.70
- 7
Imagen 4 Ultra (generate-001) 0.66
- 8
Imagen 4 (generate-001) 0.65
- 9
HunyuanImage 3.0 0.62
- 10 Ideogram 3.0 ↑1 0.53
Art
10 models we track have a score here.
- 1
Gemini 3 Pro Image (nano-banana-pro, 2K) 0.90
- 2 Grok Imagine Image 0.89
- 3 GPT-Image-1.5 0.81
- 4
Gemini 2.5 Flash Image Preview (nano-banana) ↑1 0.78
- 5 GPT-Image-1 ↑1 0.78
- 6 Recraft V4 ↓2 0.77
- 7
Imagen 4 Ultra (generate-001) 0.69
- 8
Imagen 4 (generate-001) 0.68
- 9
HunyuanImage 3.0 0.63
- 10 Ideogram 3.0 ↑1 0.52
Cartoon
10 models we track have a score here.
- 1
Gemini 3 Pro Image (nano-banana-pro, 2K) 0.82
- 2 Grok Imagine Image 0.80
- 3 GPT-Image-1.5 0.77
- 4
Gemini 2.5 Flash Image Preview (nano-banana) ↑1 0.74
- 5 Recraft V4 ↓1 0.74
- 6 GPT-Image-1 0.69
- 7
Imagen 4 Ultra (generate-001) 0.67
- 8
Imagen 4 (generate-001) 0.65
- 9
HunyuanImage 3.0 0.57
- 10 Ideogram 3.0 ↑1 0.51
Commercial design
10 models we track have a score here.
- 1 Grok Imagine Image ↑1 0.85
- 2
Gemini 3 Pro Image (nano-banana-pro, 2K) ↓1 0.84
- 3 Recraft V4 ↑1 0.81
- 4
Gemini 2.5 Flash Image Preview (nano-banana) ↑1 0.75
- 5 GPT-Image-1.5 ↓2 0.75
- 6
Imagen 4 (generate-001) ↑2 0.66
- 7 GPT-Image-1 ↓1 0.66
- 8
Imagen 4 Ultra (generate-001) ↓1 0.58
- 9
HunyuanImage 3.0 0.51
- 10 Ideogram 3.0 ↑1 0.45
Photorealistic
10 models we track have a score here.
- 1 Grok Imagine Image ↑1 0.88
- 2
Gemini 3 Pro Image (nano-banana-pro, 2K) ↓1 0.85
- 3 GPT-Image-1.5 0.83
- 4 Recraft V4 0.82
- 5
Gemini 2.5 Flash Image Preview (nano-banana) 0.78
- 6 GPT-Image-1 0.74
- 7
Imagen 4 Ultra (generate-001) 0.69
- 8
HunyuanImage 3.0 ↑1 0.67
- 9
Imagen 4 (generate-001) ↓1 0.65
- 10 Ideogram 3.0 ↑1 0.52
Portraits
10 models we track have a score here.
- 1 Recraft V4 ↑3 0.88
- 2 Grok Imagine Image 0.88
- 3
Gemini 3 Pro Image (nano-banana-pro, 2K) ↓2 0.83
- 4 GPT-Image-1.5 ↓1 0.80
- 5
Gemini 2.5 Flash Image Preview (nano-banana) 0.79
- 6
Imagen 4 Ultra (generate-001) ↑1 0.77
- 7 GPT-Image-1 ↓1 0.74
- 8
HunyuanImage 3.0 ↑1 0.70
- 9
Imagen 4 (generate-001) ↓1 0.69
- 10 Ideogram 3.0 ↑1 0.58
Text
10 models we track have a score here.
- 1
Gemini 3 Pro Image (nano-banana-pro, 2K) 0.91
- 2 Grok Imagine Image 0.88
- 3 Recraft V4 ↑1 0.82
- 4
Gemini 2.5 Flash Image Preview (nano-banana) ↑1 0.75
- 5 GPT-Image-1.5 ↓2 0.74
- 6 GPT-Image-1 0.73
- 7
Imagen 4 Ultra (generate-001) 0.66
- 8
Imagen 4 (generate-001) 0.60
- 9 Ideogram 3.0 ↑2 0.57
- 10
HunyuanImage 3.0 ↓1 0.47
Gemini-3-Pro (243 prompts)
10 models we track have a score here.
- 1
Gemini 3 Pro Image (nano-banana-pro, 2K) 0.87
- 2 Grok Imagine Image 0.85
- 3 GPT-Image-1.5 0.80
- 4 Recraft V4 0.79
- 5
Gemini 2.5 Flash Image Preview (nano-banana) 0.76
- 6 GPT-Image-1 0.73
- 7
Imagen 4 Ultra (generate-001) 0.68
- 8
Imagen 4 (generate-001) 0.67
- 9
HunyuanImage 3.0 0.62
- 10 Ideogram 3.0 ↑1 0.52
GPT-5.4 (243 prompts)
10 models we track have a score here.
- 1 Grok Imagine Image ↑1 0.80
- 2
Gemini 3 Pro Image (nano-banana-pro, 2K) ↓1 0.80
- 3 GPT-Image-1.5 0.73
- 4 Recraft V4 0.72
- 5
Gemini 2.5 Flash Image Preview (nano-banana) 0.71
- 6 GPT-Image-1 0.65
- 7
Imagen 4 Ultra (generate-001) 0.63
- 8
Imagen 4 (generate-001) 0.59
- 9
HunyuanImage 3.0 0.56
- 10 Ideogram 3.0 ↑1 0.45
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
- 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 benchmark’s paper (13 entries).
- Pass Models scored 11 models we track, plus 2 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 banyuanhao.github.io: 17 rows for models we track and 2 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.