Genspark 算账：token 单价降了 10 倍，智能体任务烧 token 太狠，总账单反而涨。省钱得盯单次任务成本，别只看单价。
今年AI智能的单位成本下降约10倍，但企业AI账单不降反升。原因并非单价上涨，而是智能体进入生产：一个会自主研究、规划、修改的智能体可消耗数百万token，而普通聊天机器人仅消耗数千。Genspark 认为，真正的优化指标应是每个成功任务的成本，而非每token成本。
Intelligence got ~10× cheaper this year. AI bills went up anyway. Not because enterprises pay mor...
Intelligence got ~10× cheaper this year. AI bills went up anyway. Not because enterprises pay more per unit. Because agents moved into production, and an agent that researches, plans, and revises its own work burns millions of tokens where a chatbot burned thousands. The real problem: teams are optimizing for cost per token when the metric that matters is cost per successful task. That’s what we're building Genspark to answer. Read more: genspark.ai/blog/the-ai-ef… 💬 0 🔄 1 ❤️ 4 👀 382 📊 1 ⚡
- Eric Jing08-06 08:45原文