查看各种 AI 大语言模型在数学推理、代码生成、创意写作及其他开放式文本任务中的综合 Elo 积分排名，数据来源于真实用户的匿名盲测投票。
基于 Bradley-Terry 统计模型计算的大模型竞技积分。数值越高代表在人类盲测对决中综合胜率越高，反映模型间相对强弱。
Competitive score based on the Bradley-Terry statistical model. Higher score = better overall win rate in human blind tests, reflecting relative model strength.
Elo 积分的完整浮点数值（精确到小数3位）。当两个模型整数分相同时，精确分才真正决定排名先后。下方 [lower, upper] 为置信区间范围。
Full floating-point Elo score (3 decimal places). When two models share the same integer score, Float Rating determines the final ranking. [lower, upper] shows the confidence interval.
衡量 Elo 积分的统计稳定程度。绿色·稳定（<5）票多可信；橙色·波动（5~12）；红色·不稳（>12）新模型票少，排名还会剧烈变化。
Measures statistical stability of the Elo score. Green·Stable (<5); Orange·Unstable (5–12); Red·Volatile (>12) — new models with fewer votes may fluctuate significantly.
真实实力有 95% 概率落在 [lower, upper] 区间内，显示为 ±N。N 越小排名越可信，票数越多 N 越小。
The model's true strength falls within [lower, upper] with 95% probability, shown as ±N. Smaller N = more reliable ranking; more votes → smaller N.
该模型参与盲测对决的总次数。票数越高，Elo 分的统计置信度越高、越不容易受单次投票影响而波动。
Total number of head-to-head comparisons this model participated in. Higher vote count = higher statistical confidence in the Elo score.
Apache/Llama：完全开源可商用；MIT：开源限制极少；Proprietary：闭源商业模型。协议决定你能否把它集成到自己的产品里。
Apache/Llama: Fully open-source, commercial use allowed; MIT: Open-source, minimal restrictions; Proprietary: Closed-source commercial model. License determines if you can integrate it into your product.
| 排名Rank | 名次区间Rank Range | 模型Model | Arena 得分Arena Score ↓ | 投票数Votes | 精确分 & 区间Float Score & CI | 方差Variance | 价格 ($/1M)Price ($/1M) | 上下文Context |
|---|---|---|---|---|---|---|---|---|
| 🥇 | 54 - 77 |
qwen2-vl-72b
阿里巴巴 · Qwen
|
1095±14.9 | 989 票votes | 1095.147 [1082.3, 1112.1] | 78.76 · 不稳Volatile | $0.90 / $0.90 | 32.8K |
没有找到匹配的模型，换个关键词试试？No matching models found. Try a different keyword?