Glean 称其企业搜索比 Claude Cowork 节省 81%

Claude Cowork used 4.4M tokens on a benchmark where Glean used 1.3M.

@glean is claiming an 81% savings on token costs over Claude Cowork while being preferred 78% of the time..

According to Glean, Cowork uses a federated MCP approach that searches each connected system individually, which can lead to overfetching results, normalizing data, resolving conflicts, and repeating reasoning loops..

Glean starts from a unified, pre-indexed, ranked view of enterprise information, reducing the context-retrieval work the model has to do.

Glean's internal benchmark puts the average task at $0.58 versus $2.98 for Cowork. The cheaper model mix contributes, and Glean also used 70% fewer tokens, which it attributes in part to its enterprise context foundations and harness design.

Glean's auto routing matches tasks to different models and reasoning levels based on the effort and capability required.

Context efficiency is starting to look just as important as model efficiency.