Drex for Search
A document search decision model for AI agents introduced by Nace AI. It enables the model to independently determine when to retrieve, how many retrieval steps to perform, and the conditions for stopping, and belongs to the retrieval strategy layer of models.
Timeline (last 90 days)
Nace AI released Drex for Search, a document search model for AI Agents.
Drex for Search achieved a score of 92.05% on the BrowseComp-Plus benchmark.
Drex for Search requires an average of only 12.6 search calls on BrowseComp-Plus.
Nace AI said Drex for Search reduced the Agent's interaction turns on OfficeQA Pro, but did not disclose the specific magnitude of the reduction.
The design philosophy of Drex for Search is to let the model decide for itself when to retrieve, how many times to retrieve, and when to stop.
Drex for Search's figures of 92.05% and 12.6 calls are vendor-reported, and the original material did not disclose the evaluation configuration, baseline comparison targets, or statistical methodology.
Drex for Search belongs to the retrieval strategy layer rather than the retrieval engine layer; it does not promise stronger recall but addresses the issues of retrieval timing and stopping conditions.
There is no public data yet on Drex for Search's performance in complex environments such as enterprise private knowledge bases, low-quality documents, and long documents.
Drex for Search's compatibility with mainstream Agent frameworks and whether it supports custom retrieval backends were not covered in the release materials.
All Facts (9)
Drex for Search's compatibility with mainstream Agent frameworks and whether it supports custom retrieval backends were not covered in the release materials.
50%UnverifiedDrex for Search achieved a score of 92.05% on the BrowseComp-Plus benchmark.
50%UnverifiedDrex for Search requires an average of only 12.6 search calls on BrowseComp-Plus.
50%UnverifiedNace AI said Drex for Search reduced the Agent's interaction turns on OfficeQA Pro, but did not disclose the specific magnitude of the reduction.
50%UnverifiedThe design philosophy of Drex for Search is to let the model decide for itself when to retrieve, how many times to retrieve, and when to stop.
50%UnverifiedDrex for Search's figures of 92.05% and 12.6 calls are vendor-reported, and the original material did not disclose the evaluation configuration, baseline comparison targets, or statistical methodology.
50%UnverifiedDrex for Search belongs to the retrieval strategy layer rather than the retrieval engine layer; it does not promise stronger recall but addresses the issues of retrieval timing and stopping conditions.
50%UnverifiedThere is no public data yet on Drex for Search's performance in complex environments such as enterprise private knowledge bases, low-quality documents, and long documents.
50%UnverifiedNace AI released Drex for Search, a document search model for AI Agents.
50%