文生图模型排行榜
本榜收录文生图(text-to-image)维度在榜的 78 个模型,按 Elo 得分降序排列,当前榜首为 GPT Image 2.5 Sunburst(OpenAI),得分 1421。该维度是 11 个维度中在榜模型最多的一项,对战方式为同一提示词下两组生成结果二选一,反映的是生成质量与提示词遵循度的综合偏好。
文生图 · 完整榜单(78 个模型)
按 Elo 得分降序| 排名 | 模型 | 厂商 / 国家 | Elo 得分 | 置信区间 | 投票数 |
|---|---|---|---|---|---|
| 1 |
GPT Image 2.5 Sunburst
|
OpenAI
|
1421 | ±13 | 3,149 |
| 2 |
GPT Image 2.5 Flare
|
OpenAI
|
1399 | ±13 | 2,856 |
| 3 |
gpt-image-2 (medium)
|
OpenAI
|
1381 | ±4 | 78,731 |
| 4 |
Mai Image 2.6
|
Microsoft AI
|
1331 | ±7 | 11,213 |
| 5 |
grok-imagine-image-2.0 (low)
|
SpaceXAI
|
1315 | ±12 | 2,681 |
| 6 | R Reve 2.1 |
Reve
|
1301 | ±8 | 7,573 |
| 7 |
Muse Image
|
Meta
|
1277 | ±6 | 26,045 |
| 8 | R Reve 2.0 |
Reve
|
1270 | ±6 | 14,632 |
| 9 |
gemini-3.1-flash-image (nano-banana-2) [web-search]
|
Google
|
1261 | ±5 | 41,957 |
| 10 | B Seedream 5.0 Pro |
Bytedance
|
1257 | ±4 | 62,443 |
| 11 |
Qwen Image 3.0 Pro
|
Alibaba
|
1254 | ±7 | 10,920 |
| 12 |
Mai Image 2.5
|
Microsoft AI
|
1254 | ±4 | 57,250 |
| 13 |
gemini-3.1-flash-lite-image (nano-banana-2-lite)
|
Google
|
1250 | ±6 | 16,371 |
| 14 |
gemini-3-pro-image-2k (nano-banana-pro)
|
Google
|
1246 | ±3 | 152,220 |
| 15 |
GPT Image 1.5 High Fidelity
|
OpenAI
|
1239 | ±3 | 151,702 |
| 16 |
gemini-3-pro-image-preview (nano-banana-pro)
|
Google
|
1232 | ±5 | 82,737 |
| 17 | I Ideogram 4.0 Quality |
Ideogram
|
1204 | ±4 | 42,169 |
| 18 |
Qwen Image 2.0 Pro 2026 06 22
|
Alibaba
|
1191 | ±6 | 12,005 |
| 19 | L Uni 1.1 Max |
Luma AI
|
1188 | ±6 | 13,469 |
| 20 |
Mai Image 2
|
Microsoft AI
|
1183 | ±5 | 49,198 |
| 21 | L Uni 1.1 |
Luma AI
|
1181 | ±5 | 33,836 |
| 22 |
Grok Imagine Image
|
SpaceXAI
|
1171 | ±3 | 236,623 |
| 23 | R Recraft V4.1 Utility Pro |
Recraft
|
1169 | ±11 | 2,516 |
| 24 |
Cosmos3-Super-Text2Image (Agentic)
|
Nvidia
|
1167 | ±8 | 7,424 |
| 25 | B Flux 2 Max |
Black Forest Labs
|
1162 | ±4 | 117,339 |
| 26 |
Grok Imagine Image Pro
|
SpaceXAI
|
1161 | ±4 | 93,639 |
| 27 | B Flux 2 Flex |
Black Forest Labs
|
1157 | ±3 | 149,288 |
| 28 | B Flux 2 Pro |
Black Forest Labs
|
1154 | ±3 | 188,083 |
| 29 | R Reve V1.5 |
Reve
|
1154 | ±4 | 36,748 |
| 30 |
Cosmos3 Super Text2Image
|
Nvidia
|
1154 | ±7 | 12,499 |
| 31 |
Hunyuan Image 3.0
|
Tencent
|
1151 | ±3 | 173,099 |
| 32 |
gemini-2.5-flash-image-preview (nano-banana)
|
Google
|
1150 | ±2 | 845,086 |
| 33 |
Imagen Ultra 4.0 Generate 001
|
Google
|
1148 | ±4 | 388,103 |
| 34 | B Seedream 4.5 |
Bytedance
|
1147 | ±3 | 259,145 |
| 35 | B Flux 2 Dev |
Black Forest Labs
|
1145 | ±4 | 73,963 |
| 36 | B Seedream 4 2k |
Bytedance
|
1140 | ±7 | 12,621 |
| 37 | B Seedream 5.0 Lite |
Bytedance
|
1138 | ±3 | 109,182 |
| 38 |
Wan2.6 T2i
|
Alibaba
|
1137 | ±3 | 201,108 |
| 39 | R Recraft V4.1 Pro |
Recraft
|
1130 | ±11 | 2,680 |
| 40 |
Imagen 4.0 Generate 001
|
Google
|
1129 | ±3 | 531,811 |
| 41 |
Qwen Image 2512
|
Alibaba
|
1125 | ±4 | 92,541 |
| 42 | K Krea 2 Medium |
Krea
|
1123 | ±5 | 40,205 |
| 43 |
Wan2.5 T2i Preview
|
Alibaba
|
1118 | ±3 | 248,411 |
| 44 |
Hidream o1 Image
|
HiDream
|
1117 | ±4 | 48,201 |
| 45 | B Seedream 4 Fal |
Bytedance
|
1116 | ±7 | 11,870 |
| 46 |
GPT Image 1
|
OpenAI
|
1115 | ±3 | 264,897 |
| 47 | R Recraft V4 |
Recraft
|
1114 | ±3 | 114,295 |
| 48 | B Seedream 4 High Res Fal |
Bytedance
|
1113 | ±3 | 175,361 |
| 49 | K Krea 2 Turbo |
Krea
|
1110 | ±5 | 33,237 |
| 50 |
GPT Image 1 Mini
|
OpenAI
|
1109 | ±3 | 163,843 |
| 51 | K Krea 2 Large |
Krea
|
1108 | ±5 | 39,614 |
| 52 |
Wan2.7 Image Pro
|
Alibaba
|
1103 | ±5 | 28,729 |
| 53 |
Wan2.7 Image
|
Alibaba
|
1100 | ±5 | 29,089 |
| 54 |
Mai Image 1
|
Microsoft AI
|
1093 | ±4 | 94,562 |
| 55 |
Z Image Turbo
|
Alibaba
|
1084 | ±5 | 22,251 |
| 56 | B Seedream 3 |
Bytedance
|
1082 | ±5 | 36,920 |
| 57 | B Flux 1 Kontext Max |
Black Forest Labs
|
1074 | ±3 | 65,618 |
| 58 | B Flux 2 Klein 9b |
Black Forest Labs
|
1070 | ±3 | 142,642 |
| 59 |
Qwen Image Prompt Extend
|
Alibaba
|
1061 | ±3 | 705,193 |
| 60 | B Flux 1 Kontext Pro |
Black Forest Labs
|
1059 | ±3 | 330,851 |
| 61 |
Imagen 3.0 Generate 002
|
Google
|
1058 | ±3 | 360,152 |
| 62 |
Qwen Image
|
Alibaba
|
1057 | ±3 | 84,717 |
| 63 | I Ideogram V3 Quality |
Ideogram
|
1049 | ±4 | 115,413 |
| 64 | L Photon |
Luma AI
|
1035 | ±4 | 127,530 |
| 65 | AI P Image |
|
1034 | ±4 | 103,848 |
| 66 | B Flux 2 Klein 4b |
Black Forest Labs
|
1030 | ±3 | 144,483 |
| 67 | R Runway Gen4 |
Runway
|
1025 | ±5 | 51,755 |
| 68 | R Recraft V3 |
Recraft
|
1021 | ±4 | 191,836 |
| 69 | B Flux 1.1 Pro |
Black Forest Labs
|
1016 | ±4 | 70,460 |
| 70 | L Lucid Origin |
Leonardo AI
|
1013 | ±3 | 285,042 |
| 71 | I Ideogram V2 |
Ideogram
|
1013 | ±4 | 72,090 |
| 72 | Z GLM Image |
Z.ai
|
1010 | ±9 | 4,624 |
| 73 |
Gemini 2.0 Flash Preview Image Generation
|
Google
|
975 | ±3 | 257,085 |
| 74 | B Flux 1 Dev Fp8 |
Black Forest Labs
|
969 | ±4 | 49,221 |
| 75 |
Dall E 3
|
OpenAI
|
968 | ±4 | 239,060 |
| 76 | B Flux 1 Kontext Dev |
Black Forest Labs
|
940 | ±4 | 215,336 |
| 77 | AI Stable Diffusion V35 Large |
|
938 | ±5 | 23,396 |
| 78 | B Bagel |
Bytedance
|
898 | ±6 | 12,375 |
本榜国家与厂商分布
文生图维度:78 个模型,来自 19 家厂商(按数量列前 8 家)
本维度在榜 78 个模型,来自 19 家厂商,厂商分布较分散,数量最多的 Black Forest Labs 也只占 14%。国家分布上,中国厂商 20 席、美国厂商 58 席;中国厂商占本维度 26%,与全部 11 个维度的整体水平(24%)大致相当。榜首 GPT Image 2.5 Sunburst(OpenAI)得分 1421,领先第 2 名 22 分;其中 5 个模型置信区间在 ±11 以上(对战样本偏少),名次仍有变动空间。
榜单说明
排序依据:按 Elo 得分降序。得分由真实用户在不知模型身份的前提下对两组回答二选一(盲评对战),经 Elo 算法累积计算, 代表相对胜率而非绝对能力值。置信区间越小表示对战样本越充足;投票数为该模型的累计有效对战场次。 不同维度之间的分数不可横向比较(例如编程维度榜首约 1800 分、文本对话维度榜首约 1506 分,并不代表编程模型更强)。 国家标识依据模型归属厂商判定,同一厂商的多个版本归为同一国家。 数据来源:arena-ai-leaderboards 公开接口,最近抓取时间 2026-09-23 00:10:00,本站每日抓取快照并留存历史。 榜单仅反映真实用户偏好,不构成技术评测结论。 相关榜单:社区榜单(工具热度榜与开源模型下载榜)、 模型库(按厂商 / 参数 / 下载量筛选)。