AI 导读
企业 AI 昂贵的地方，往往是对任务用了过多的智能。 Glean 正在把这个决策移入运行时。 - token 成本降低 81%，相比 Claude Cowork 有 78% 的情况下更受偏好。 - Glean 每次查询 $0.58，对比 $2.98。 依据是他们覆盖 180+ 企业任务的基准测试。 我喜欢他们这个基准测试的地方在于，它并不是真正的模型对模型。它是固定模型推理 vs. 上下文 + 路由作为一个系统。
正文
The expensive part of enterprise AI is often using too much intelligence for the task.
Glean is moving that decision into the runtime.
- 81% lower token cost and preferred 78% of the time vs Claude Cowork.
- $0.58 per query for Glean against $2.98.
per their benchmark across 180+ enterprise tasks
What I like about their benchmark is that it isn't really model vs. model. It's fixed-model inference vs. context + routing as a system.
来源：@rohanpaul_ai · x.com