📄 中文摘要
当前有超过400个AI模型可通过API使用，作者在过去两个月内测试了359个模型，使用真实的提示、测量实际输出并追踪真实成本。结果显示，没有单一模型在所有类别中都表现最佳。文章还提供了2026年2月的定价表，列出了每百万输入令牌的成本，展示了不同模型之间的价格差异，部分模型的成本差异达到100倍。通过这些数据，用户可以更明智地选择适合的AI模型。
我测试了359个AI模型——哪些值得付费
出处: I Tested 359 AI Models — Here’s Which Ones Are Actually Worth Paying For
发布: 2026年2月18日
📄 English Summary
I Tested 359 AI Models — Here’s Which Ones Are Actually Worth Paying For
There are over 400 AI models available via API, and the author tested 359 of them over the past two months using real prompts, measuring actual outputs, and tracking real costs. The findings reveal that no single model excels in every category. The article also includes a pricing table from February 2026, detailing the costs per million input tokens, highlighting a staggering 100x difference in pricing among various models. This data enables users to make informed decisions when selecting suitable AI models.