Artificials
Compare models
IMAGE MODELS

The best AI image models, ranked by people.

Arena shows people two images made from the same prompt, without naming the models, and asks which is better. Millions of those votes decide these rankings.

At a glance

What people’s votes say about today’s image models.

Leader

GPT Image 2.5 Sunburst leads text to image at 1425, 28 points ahead of GPT Image 2.5 Flare. Their 95% ranges do not overlap: a clear lead.

Open weights

The best open-weights model, Qwen Image 2.1 (Qwen Research License), ranks #18 at 1223, 202 points behind the leader.

Categories

GPT Image 2.5 Sunburst also leads all 7 categories.

New

New on the board in the last month: GPT Image 2.5 Sunburst, GPT Image 2.5 Flare, Grok Imagine Image 2.0 (Canvas), MAI Image 2.6, Grok Imagine Image 2.0 (20260801), Qwen Image 2.1 and Recraft V4.1 Flash.

Editing

GPT Image 2.5 Sunburst leads image editing at 1524, out of 58 models.

Text to image

81 models, ranked by blind votes on the same prompt.

Text to image: overall

Arena score, higher is better. Pick a kind of image, or show only models with downloadable weights.

  1. Rank 1GPT Image 2.5 SunburstOpenAI, 16,652 votesNewArena score 142595% range 1418 to 1432
  2. Rank 2GPT Image 2.5 FlareOpenAI, 15,792 votesNewArena score 139795% range 1390 to 1404
  3. Rank 3GPT Image 2 (medium)OpenAI, 92,549 votesArena score 138495% range 1379 to 1388
  4. Rank 4Grok Imagine Image 2.0 (Canvas)xAI, 3,070 votesNewArena score 133695% range 1324 to 1347
  5. Rank 5MAI Image 2.6Microsoft AI, 23,918 votesNewArena score 133395% range 1328 to 1339
  6. Rank 6Reve 2.1Reve, 8,250 votesArena score 130295% range 1294 to 1310
  7. Rank 7Grok Imagine Image 2.0 (20260801)xAI, 13,278 votesNewArena score 129795% range 1290 to 1303
  8. Rank 8Muse ImageMeta, 41,071 votesArena score 127495% range 1269 to 1279
  9. Rank 9Reve 2.0Reve, 15,774 votesArena score 126995% range 1263 to 1275
  10. Rank 10Gemini 3.1 Flash Image (Nano Banana 2), web searchGoogle, 59,301 votesArena score 126195% range 1257 to 1265
  11. Rank 11Seedream 5.0 ProByteDance, 93,111 votesArena score 125695% range 1252 to 1260
  12. Rank 12Qwen Image 3.0 ProAlibaba, 11,443 votesArena score 125595% range 1249 to 1262
  13. Rank 13MAI Image 2.5Microsoft AI, 61,268 votesArena score 125495% range 1250 to 1259
  14. Rank 14Gemini 3.1 Flash Lite Image (Nano Banana 2 Lite)Google, 17,808 votesArena score 125095% range 1244 to 1256
  15. Rank 15Gemini 3 Pro Image 2K (Nano Banana Pro)Google, 174,535 votesArena score 124895% range 1245 to 1251
  16. Rank 16GPT Image 1.5 High FidelityOpenAI, 169,194 votesArena score 123895% range 1235 to 1241
  17. Rank 17Gemini 3 Pro Image Preview (Nano Banana Pro)Google, 86,464 votesArena score 123495% range 1229 to 1239
  18. Rank 18Qwen Image 2.1Alibaba, 10,231 votesOpen weightsNewArena score 122395% range 1216 to 1229
  19. Rank 19Ideogram 4.0 QualityIdeogram, 59,356 votesOpen weightsArena score 120595% range 1201 to 1209
  20. Rank 20Qwen Image 2.0 Pro (2026-06-22)Alibaba, 13,099 votesArena score 119295% range 1186 to 1198

Dots mark each model’s Arena score; the line is its 95% range. Where two ranges overlap, the order between those models is not settled. Arena (CC BY 4.0).

The best model for each kind of image

The top model from each of the three leading makers, with its rank, from the votes on prompts of that kind.

Overall

  1. Rank 1GPT Image 2.5 Sunburst1425
  2. Rank 4Grok Imagine Image 2.0 (Canvas)1336
  3. Rank 5MAI Image 2.61333

Photorealistic

  1. Rank 1GPT Image 2.5 Sunburst1430
  2. Rank 4MAI Image 2.61331
  3. Rank 5Grok Imagine Image 2.0 (Canvas)1327

Portraits

  1. Rank 1GPT Image 2.5 Sunburst1467
  2. Rank 4Grok Imagine Image 2.0 (Canvas)1358
  3. Rank 5MAI Image 2.61347

Art

  1. Rank 1GPT Image 2.5 Sunburst1395
  2. Rank 4Grok Imagine Image 2.0 (Canvas)1339
  3. Rank 5MAI Image 2.61336

Cartoon

  1. Rank 1GPT Image 2.5 Sunburst1445
  2. Rank 4Grok Imagine Image 2.0 (Canvas)1373
  3. Rank 5MAI Image 2.61350

3D modeling

  1. Rank 1GPT Image 2.5 Sunburst1414
  2. Rank 4MAI Image 2.61327
  3. Rank 5Grok Imagine Image 2.0 (Canvas)1322

Commercial design

  1. Rank 1GPT Image 2.5 Sunburst1420
  2. Rank 4Grok Imagine Image 2.0 (Canvas)1348
  3. Rank 5MAI Image 2.61340

Text in images

  1. Rank 1GPT Image 2.5 Sunburst1471
  2. Rank 4MAI Image 2.61371
  3. Rank 5Grok Imagine Image 2.0 (Canvas)1357

Makers

Where each company’s best image model stands.

Each maker’s best text-to-image model

The highest-scoring model from each of the 12 leading makers.

  1. OpenAIGPT Image 2.5 SunburstArena score 142595% range 1418 to 1432
  2. xAIGrok Imagine Image 2.0 (Canvas)Arena score 133695% range 1324 to 1347
  3. Microsoft AIMAI Image 2.6Arena score 133395% range 1328 to 1339
  4. ReveReve 2.1Arena score 130295% range 1294 to 1310
  5. MetaMuse ImageArena score 127495% range 1269 to 1279
  6. GoogleGemini 3.1 Flash Image (Nano Banana 2), web searchArena score 126195% range 1257 to 1265
  7. ByteDanceSeedream 5.0 ProArena score 125695% range 1252 to 1260
  8. AlibabaQwen Image 3.0 ProArena score 125595% range 1249 to 1262
  9. IdeogramIdeogram 4.0 QualityArena score 120595% range 1201 to 1209
  10. Luma AIUNI 1.1 MaxArena score 118995% range 1183 to 1196
  11. RecraftRecraft V4.1 Utility ProArena score 116995% range 1158 to 1180
  12. NVIDIACosmos3-Super-Text2Image (Agentic)Arena score 116595% range 1158 to 1172

Dots mark the Arena score; the line is its 95% range. Arena (CC BY 4.0).

Progress

How the leading makers’ best scores have moved, month by month.

Best text-to-image score, by maker

Each point is a maker’s best model at the end of a month. Select a maker in the legend to hide or show it.

Arena score. Use the arrow keys to move between points.

Best text-to-image Arena score by maker, month by month
SeriesDateModelArena score
OpenAIJan 28, 2025DALL·E 3978
OpenAIFeb 27, 2025DALL·E 3978
OpenAIMar 30, 2025DALL·E 3978
OpenAIApr 23, 2025DALL·E 3978
OpenAIMay 22, 2025DALL·E 3978
OpenAIJun 24, 2025GPT Image 11135
OpenAIJul 28, 2025GPT Image 11137
OpenAIAug 28, 2025GPT Image 11124
OpenAISep 30, 2025GPT Image 11127
OpenAIOct 17, 2025GPT Image 11124
OpenAINov 21, 2025GPT Image 11122
OpenAIDec 16, 2025GPT Image 1.5 High Fidelity1264
OpenAIJan 29, 2026GPT Image 1.5 High Fidelity1236
OpenAIFeb 26, 2026GPT Image 1.5 High Fidelity1248
OpenAIMar 26, 2026GPT Image 1.5 High Fidelity1244
OpenAIApr 23, 2026GPT Image 2 (medium)1508
OpenAIMay 26, 2026GPT Image 2 (medium)1388
OpenAIJun 30, 2026GPT Image 2 (medium)1387
OpenAIJul 10, 2026GPT Image 2 (medium)1385
OpenAIAug 7, 2026GPT Image 2 (medium)1381
OpenAISep 24, 2026GPT Image 2.5 Sunburst1424
OpenAIOct 5, 2026GPT Image 2.5 Sunburst1425
xAIJan 29, 2026Grok Imagine Image Pro1166
xAIMar 26, 2026Grok Imagine Image1173
xAIApr 23, 2026Grok Imagine Image1170
xAIMay 26, 2026Grok Imagine Image Quality1228
xAIJun 30, 2026Grok Imagine Image Quality1231
xAIJul 10, 2026Grok Imagine Image Quality1229
xAIAug 7, 2026Grok Imagine Image Quality1228
xAISep 24, 2026Grok Imagine Image 2.0 (low)1301
xAIOct 5, 2026Grok Imagine Image 2.0 (Canvas)1336
Microsoft AIOct 17, 2025MAI Image 11090
Microsoft AINov 21, 2025MAI Image 11090
Microsoft AIDec 16, 2025MAI Image 11092
Microsoft AIJan 29, 2026MAI Image 11095
Microsoft AIFeb 26, 2026MAI Image 11094
Microsoft AIMar 26, 2026MAI Image 21190
Microsoft AIApr 23, 2026MAI Image 21183
Microsoft AIMay 26, 2026MAI Image 2.5 Preview1254
Microsoft AIJun 30, 2026MAI Image 2.51257
Microsoft AIJul 10, 2026MAI Image 2.51257
Microsoft AIAug 7, 2026MAI Image 2.51256
Microsoft AISep 24, 2026MAI Image 2.61335
Microsoft AIOct 5, 2026MAI Image 2.61333
ReveFeb 26, 2026Reve V1.51178
ReveMar 26, 2026Reve V1.51177
ReveApr 23, 2026Reve V1.51177
ReveMay 26, 2026Reve V1.51156
ReveJun 30, 2026Reve 2.01273
ReveJul 10, 2026Reve 2.11302
ReveAug 7, 2026Reve 2.11302
ReveSep 24, 2026Reve 2.11302
ReveOct 5, 2026Reve 2.11302
MetaJul 10, 2026Muse Image1280
MetaAug 7, 2026Muse Image1283
MetaSep 24, 2026Muse Image1276
MetaOct 5, 2026Muse Image1274
GoogleJan 28, 2025Imagen 3.0 (002)1105
GoogleFeb 27, 2025Imagen 3.0 (002)1090
GoogleMar 30, 2025Imagen 3.0 (002)1088
GoogleApr 23, 2025Imagen 3.0 (002)1086
GoogleMay 22, 2025Imagen 3.0 (002)1084
GoogleJun 24, 2025Imagen 3.0 (002)1081
GoogleJul 28, 2025Imagen Ultra 4.0 (001)1114
GoogleAug 28, 2025Gemini 2.5 Flash Image Preview (Nano Banana)1145
GoogleSep 30, 2025Gemini 2.5 Flash Image Preview (Nano Banana)1150
GoogleOct 17, 2025Gemini 2.5 Flash Image Preview (Nano Banana)1153
GoogleNov 21, 2025Gemini 3 Pro Image Preview (Nano Banana Pro)1242
GoogleDec 16, 2025Gemini 3 Pro Image Preview 2K (Nano Banana Pro)1235
GoogleJan 29, 2026Gemini 3 Pro Image Preview 2K (Nano Banana Pro)1228
GoogleFeb 26, 2026Gemini 3.1 Flash Image Preview (Nano Banana 2), web search1280
GoogleMar 26, 2026Gemini 3.1 Flash Image Preview (Nano Banana 2), web search1266
GoogleApr 23, 2026Gemini 3.1 Flash Image Preview (Nano Banana 2), web search1271
GoogleMay 26, 2026Gemini 3.1 Flash Image Preview (Nano Banana 2), web search1271
GoogleJun 30, 2026Gemini 3.1 Flash Image Preview (Nano Banana 2), web search1270
GoogleJul 10, 2026Gemini 3.1 Flash Image (Nano Banana 2), web search1261
GoogleAug 7, 2026Gemini 3.1 Flash Image (Nano Banana 2), web search1263
GoogleSep 24, 2026Gemini 3.1 Flash Image (Nano Banana 2), web search1261
GoogleOct 5, 2026Gemini 3.1 Flash Image (Nano Banana 2), web search1261

Arena recomputes every score at each publication, and the scale shifts as models join, so read the lines for who led and when more than for exact gains. Arena (CC BY 4.0).

Image editing

58 models changing a picture on instruction, ranked the same way.

Image editing: overall

Arena score for edits of the same image and instruction, higher is better.

  1. Rank 1GPT Image 2.5 SunburstOpenAI, 58,172 votesNewArena score 152495% range 1518 to 1529
  2. Rank 2GPT Image 2.5 FlareOpenAI, 55,745 votesNewArena score 148195% range 1476 to 1486
  3. Rank 3GPT Image 2 (medium)OpenAI, 304,260 votesArena score 146295% range 1459 to 1465
  4. Rank 4Grok Imagine Image 2.0 (Canvas)xAI, 10,821 votesNewArena score 143895% range 1429 to 1447
  5. Rank 5MAI Image 2.6Microsoft AI, 50,147 votesNewArena score 142795% range 1422 to 1432
  6. Rank 6Grok Imagine Image 2.0 (20260801)xAI, 48,559 votesNewArena score 142695% range 1420 to 1431
  7. Rank 7Muse ImageMeta, 141,017 votesArena score 140395% range 1398 to 1407
  8. Rank 8MAI Image 2.5Microsoft AI, 190,078 votesArena score 140195% range 1397 to 1405
  9. Rank 9Seedream 5.0 ProByteDance, 281,170 votesArena score 139495% range 1391 to 1397
  10. Rank 10Grok Imagine Image Quality (20260519)xAI, 38,195 votesArena score 139195% range 1386 to 1397
  11. Rank 11Gemini 3 Pro Image 2K (Nano Banana Pro)Google, 649,919 votesArena score 139095% range 1387 to 1393
  12. Rank 12ChatGPT Image Latest High Fidelity (20251216)OpenAI, 627,869 votesArena score 138995% range 1386 to 1391
  13. Rank 13Gemini 3.1 Flash Image (Nano Banana 2), web searchGoogle, 228,014 votesArena score 138795% range 1384 to 1390
  14. Rank 14Gemini 3 Pro Image Preview (Nano Banana Pro)Google, 540,775 votesArena score 138695% range 1383 to 1389
  15. Rank 15Reve 2.1Reve, 20,794 votesArena score 137495% range 1368 to 1380
  16. Rank 16GPT Image 1.5 High FidelityOpenAI, 652,679 votesArena score 137095% range 1367 to 1372
  17. Rank 17Qwen Image 2.1Alibaba, 21,114 votesOpen weightsNewArena score 136995% range 1363 to 1374
  18. Rank 18Reve 2.0Reve, 19,263 votesArena score 135995% range 1352 to 1365
  19. Rank 19Ideogram 4.5Ideogram, 17,351 votesNewArena score 134795% range 1342 to 1353
  20. Rank 20UNI 1.1 MaxLuma AI, 43,441 votesArena score 133495% range 1329 to 1339

Dots mark each model’s Arena score; the line is its 95% range. Where two ranges overlap, the order between those models is not settled. Arena (CC BY 4.0).

An automated test

Votes measure what people prefer. T2I-CoReBench checks something else: whether a model draws exactly what a complex prompt asks for.

Composition and reasoning

T2I-CoReBench score, higher is better. Its models and versions can differ from Arena’s.
  1. Nano Banana 285.3
  2. Nano Banana Pro85.0
  3. GPT-Image-1.578.2
  4. Seedream 4.577.0
  5. Imagen 4 Ultra76.1
  6. Nano Banana75.9
  7. Seedream 4.075.3
  8. GPT-Image72.6
  9. Imagen 470.0
  10. FLUX.2-dev64.4
  11. Qwen-Image-251262.4
  12. Z-Image61.1

Frequently asked questions

Answered from the data on this page.

Which is the best AI image generator?

On Arena’s text-to-image board, GPT Image 2.5 Sunburst scores highest at 1425, from 16,652 votes, ahead of GPT Image 2.5 Flare (1397) and GPT Image 2 (medium) (1384). The best model for your work can differ by category, such as photos, portraits or text in images.

What is the best open-weights image model?

Qwen Image 2.1 ranks highest among models with downloadable weights, #18 overall at 1223, under the Qwen Research License. Next is Ideogram 4.0 Quality at #19. Check each licence: several allow only non-commercial use.

Which AI makes the most realistic photos?

GPT Image 2.5 Sunburst leads Arena’s photorealistic category at 1430. Categories use only the votes on prompts of that kind.

Which AI is best at putting text in images?

GPT Image 2.5 Sunburst leads Arena’s text in images category at 1471. Categories use only the votes on prompts of that kind.

Which AI is best for portraits?

GPT Image 2.5 Sunburst leads Arena’s portraits category at 1467. Categories use only the votes on prompts of that kind.

Which AI is best at editing images?

GPT Image 2.5 Sunburst leads Arena’s image-editing board at 1524, ahead of GPT Image 2.5 Flare (1481). People compare two edits of the same image and instruction.

How are the image models ranked?

Arena shows a visitor two images made from the same prompt by two anonymous models and asks which is better. Only after voting are the models named. Millions of these votes give each model a score, on a scale like chess ratings: a higher score wins more often. The 95% range shows how sure the score is; where two ranges overlap, their order is not settled.

Does Artificials run these votes?

No. The votes and scores come from Arena’s published leaderboard dataset. Artificials tidies the model names, matches models to their makers and builds the charts, category summaries and progress lines from that data.

Sources and method

Artificials does not run these votes or tests. Scores, ranges, votes and licences come from the sources below; Artificials tidies the model names for reading, matches each model to its maker, and builds the category summaries, maker comparison and progress lines. Categories that cover fewer than twenty models are left out.