AI Subscription Points Are Hard to Compare—and Often Hard to Explain
AI subscriptions and agent products increasingly sell memberships, credits or point packages, but consumers cannot reliably compare what those allowances will complete across platforms. 36氪, citing IT时报, documents user reports of large deductions for failed, retried or unexpectedly expensive tasks, while noting that billing records often do not explain execution steps, retries or the source of each deduction. Some platforms offer pre-task cost estimates, but estimates may differ from final charges, and failure compensation, expiry and support rules vary widely.
The investigation highlights a consumer-facing gap between AI pricing and usable work: without comparable point definitions, execution details and transparent failure handling, users cannot reliably estimate the cost of completing a task.
一个小时不到，近6万积分。
扣子用户王明（化名）最近尝试AI视频制作时，遇到一笔“意外支出”。此前他主要用扣子运行Agent辅助工作，开通个人高阶版会员后，每月有9.9万积分，在他看来足够日常使用，甚至经常用不完。但当他开始尝试视频生成，情况变了。“就生成了三四段十几秒的视频，大部分时间在等待，就花掉近6万积分。”他粗略估算，这相当于一小时花掉近60元，“我这才刚刚开始尝试，如果需要不断调整提示词、迭代优化视频，感觉几百元的会员费都不够。”
于是他开始精打细算：先完成日常任务，到了月底如果还有多余积分，才敢尝试新功能或视频制作。
用户购买AI时，明明是冲着完成任务的能力，最终结算时看到的却往往只有一串积分数字。任务执行了多少步骤、为什么需要反复尝试、积分究竟花在了哪里，用户未必能从账单中找到答案。
#扣子#千问办公#Qoder#WorkBuddy#TeleAgent#ChatGPT#Claude Code#OpenAI API#Anthropic API#阿里云百炼#千问AI#天猫AI空间站