大模型排行榜 · AI 模型榜单
一页聚合主流 AI 基准与榜单——arena.ai 竞技场 Elo、Artificial Analysis 智能指数与语音竞技场、OpenRouter 用量——覆盖对话、代码、TTS、语音识别、图片、视频与智能体。
| 排名 | 模型 | 厂商 | Elo ? | 票数 ? | 输入 ? | 输出 | 缓存 ? |
|---|---|---|---|---|---|---|---|
| 1 |  gpt-image-2.5-sunburst | OpenAI | 1524 ±5 | 59,825 | — | — | — |
| 2 | gpt-image-2.5-flare | OpenAI | 1481 ±5 | 57,408 | — | — | — |
| 3 | gpt-image-2 (medium) | OpenAI | 1462 ±3 | 305,825 | — | — | — |
| 4 |  grok-imagine-image-2.0 (canvas) | SpaceXAI | 1439 ±9 | 11,589 | — | — | — |
| 5 | Microsoft AI | 1428 ±5 | 51,384 | — | — | — | |
| 6 |  gemini-nano-banana-2.1 | 1428 ±6 | 12,985 | 1.50 | 7.50 | — | |
| 7 | grok-imagine-image-2.0 (20260801) | SpaceXAI | 1425 ±5 | 50,427 | — | — | — |
| 8 |  muse-image | Meta | 1403 ±4 | 142,644 | — | — | — |
| 9 | Microsoft AI | 1401 ±4 | 190,081 | — | — | — | |
| 10 |  seedream-5.0-pro | Bytedance | 1394 ±3 | 282,865 | — | — | — |
| 11 | grok-imagine-image-quality (20260519) | SpaceXAI | 1391 ±6 | 38,195 | — | — | — |
| 12 | gemini-3-pro-image-2k (nano-banana-pro) | 1390 ±3 | 651,674 | — | — | — | |
| 13 | chatgpt-image-latest-high-fidelity (20251216) | OpenAI | 1389 ±3 | 629,464 | — | — | — |
| 14 | gemini-3.1-flash-image (nano-banana-2) [web-search] | 1387 ±3 | 229,782 | 0.500 | 3.00 | — | |
| 15 | gemini-3-pro-image-preview (nano-banana-pro) | 1386 ±3 | 540,775 | 2.00 | 12.0 | 0.200 | |
| 16 |  reve-2.1 | Reve | 1374 ±6 | 20,792 | — | — | — |
| 17 | gpt-image-1.5-high-fidelity | OpenAI | 1370 ±3 | 654,249 | — | — | — |
| 18 |  qwen-image-2.1 | Alibaba | 1369 ±6 | 21,941 | — | — | — |
| 19 | reve-2.0 | Reve | 1359 ±6 | 19,265 | — | — | — |
| 20 |  ideogram-4.5 | Ideogram | 1347 ±5 | 19,318 | — | — | — |
| 21 |  uni-1.1-max | Luma AI | 1334 ±5 | 43,446 | — | — | — |
| 22 | grok-imagine-image | SpaceXAI | 1329 ±2 | 756,681 | — | — | — |
| 23 | uni-1.1 | Luma AI | 1315 ±3 | 188,834 | — | — | — |
| 24 | gemini-3.1-flash-lite-image (nano-banana-2-lite) | 1314 ±5 | 65,669 | 0.250 | 1.50 | — | |
| 25 | qwen-image-2.0-pro-2026-06-22 | Alibaba | 1303 ±5 | 56,723 | — | — | — |
| 26 |  hunyuan-image-3.0-instruct | Tencent | 1303 ±3 | 331,423 | — | — | — |
| 27 | wan2.7-image-pro | Alibaba | 1303 ±4 | 43,974 | — | — | — |
| 28 | seedream-4.5 | Bytedance | 1302 ±2 | 1,294,217 | — | — | — |
| 29 | wan2.7-image | Alibaba | 1301 ±4 | 44,796 | — | — | — |
| 30 | seedream-5.0-lite | Bytedance | 1294 ±3 | 459,744 | — | — | — |
| 31 | gemini-2.5-flash-image-preview (nano-banana) | 1293 ±2 | 11,383,435 | 0.300 | 2.50 | 0.030 | |
| 32 | seedream-4-2k | Bytedance | 1270 ±7 | 210,059 | — | — | — |
| 33 | reve-v1.1 | Reve | 1262 ±3 | 672,625 | — | — | — |
| 34 |  flux-2-max | Black Forest Labs | 1262 ±3 | 375,861 | — | — | — |
| 35 |  kling-image-o1 | KlingAI | 1251 ±4 | 143,639 | — | — | — |
| 36 | flux-2-pro | Black Forest Labs | 1245 ±3 | 637,986 | — | — | — |
| 37 | qwen-image-edit | Alibaba | 1242 ±3 | 2,021,952 | — | — | — |
| 38 | reve-v1 | Reve | 1235 ±5 | 395,313 | — | — | — |
| 39 | qwen-image-edit-2511 | Alibaba | 1234 ±3 | 485,078 | — | — | — |
| 40 | wan2.6-image | Alibaba | 1231 ±3 | 677,385 | — | — | — |
| 41 | flux-2-flex | Black Forest Labs | 1225 ±3 | 443,912 | — | — | — |
| 42 | flux-2-klein-9b | Black Forest Labs | 1225 ±3 | 540,742 | — | — | — |
| 43 | flux-2-dev | Black Forest Labs | 1225 ±4 | 254,281 | — | — | — |
| 44 | seedream-4-high-res-fal | Bytedance | 1217 ±3 | 1,229,850 | — | — | — |
| 45 | p-image-edit | — | 1211 ±3 | 359,880 | — | — | — |
| 46 | seedream-4-fal | Bytedance | 1210 ±6 | 152,113 | — | — | — |
| 47 | reve-v1.1-fast | Reve | 1207 ±3 | 553,317 | — | — | — |
| 48 | reve-edit-fast | Reve | 1198 ±4 | 230,283 | — | — | — |
| 49 | flux-2-klein-4b | Black Forest Labs | 1187 ±3 | 541,304 | — | — | — |
| 50 | wan2.5-i2i-preview | Alibaba | 1181 ±3 | 723,697 | — | — | — |
| 51 | flux-1-kontext-max | Black Forest Labs | 1181 ±3 | 390,852 | — | — | — |
| 52 | flux-1-kontext-pro | Black Forest Labs | 1176 ±3 | 6,404,567 | — | — | — |
| 53 | flux-1-kontext-dev | Black Forest Labs | 1149 ±3 | 3,639,808 | — | — | — |
| 54 | seededit-3.0 | Bytedance | 1139 ±3 | 4,918,008 | — | — | — |
| 55 | gpt-image-1 | OpenAI | 1139 ±3 | 2,869,941 | — | — | — |
| 56 | gpt-image-1-mini | OpenAI | 1124 ±3 | 682,914 | — | — | — |
| 57 | gemini-2.0-flash-preview-image-generation | 1081 ±2 | 4,942,834 | — | — | — | |
| 58 | bagel | Bytedance | 1027 ±6 | 13,488 | — | — | — |
| 59 |  step1x-edit | StepFun | 998 ±4 | 155,418 | — | — | — |
数据来源: arena.ai · 更新于 Oct 6, 2026 · PandaNpc 更新于 2026年10月13日 UTC 06:00查看完整榜单 →
指标怎么读
大模型排行榜告诉你「人更喜欢谁」，AI 基准测试告诉你「它客观上能做什么」，用量数据告诉你「开发者真的在用谁」。没有哪一张榜能单独回答「最好的 AI 模型是哪个」，所以这一页把三类都聚合进来——arena.ai 竞技场 Elo、Artificial Analysis 智能指数与语音竞技场、OpenRouter token 占比——并每日刷新。
arena.ai（原 LMArena）让用户在不知道模型身份的情况下对同一问题的两份回答二选一，再用 Elo 算法推算等级分。它衡量的是「人更喜欢哪个」，不等同于客观能力——擅长讨人喜欢的风格会占优，而严谨但啰嗦的回答可能吃亏。
票数是该模型累计参与的对战场次。新模型刚上榜时票数少、置信区间宽，排名容易波动，看排名时应连同 ± 区间一起判断。
Agent 能力榜的口径与其他榜不同：它统计的是真实 Agent 会话，而非盲投偏好。主指标「净改进」是百分比形式的提升幅度，与 Elo 不在同一量纲，不能跨榜比较。arena.ai 还给出确认成功率、可引导性、Bash 恢复等五个维度，可在完整榜单查看。
用量热度榜来自 OpenRouter，统计的是真实 API 调用的 token 份额，代表开发者用脚投票的结果，与主观偏好榜互为补充。
语音榜来自 Artificial Analysis：TTS 采用人类盲听的 Speech Arena Elo；ASR 采用 AA-WER v2，数值越低越好；原生语音对语音榜则等权综合 Big Bench Audio、Full Duplex Bench 和 τ-Voice，数值越高越好。不同榜的分数不能横向比较。来源 artificialanalysis.ai。
智能指数榜来自 Artificial Analysis：把多项公开基准测试综合成一个 0-100 的 Intelligence Index，衡量的是客观能力而非主观偏好，正好与 arena.ai 的人类盲投互补。数据来源 artificialanalysis.ai。