Nace AI launches drex for search, achieving 92.05% accuracy on browsecomp-plus

Nace AI has launched Drex for Search, a retrieval layer designed to enhance document search for AI agents. The tool integrates Nace's Drex decision models, enabling agents like Claude Code and Codex to find evidence efficiently. In evaluations, Drex for Search achieved 92.05% accuracy on BrowseComp-Plus using GPT-5, with 12.6 search calls per question. The system is framed as a series of decisions, with Drex scoring options in a single forward pass, making it cost-effective for each agent step. Nace AI tested Drex for Search across nine agent-model pairs on OfficeQA Pro, a 133-question benchmark, showing 6% higher accuracy, 62% fewer steps, and 56% fewer tokens. On BrowseComp-Plus, Drex for Search scored 77.5% R@1 with Qwen 27B, outperforming Claude Opus 5. The tool also achieved 94.7% accuracy on MADQA with Gemini 3.5 Flash, and ranked #1 on EnterpriseRAG-Bench with an overall score of 88.22, surpassing OpenAI File Search by 27.2 points. Drex for Search is now available via the Nace API Console, with access through four methods. The release highlights Nace AI's advancements in search efficiency and accuracy, positioning it as a competitive player in the AI search landscape.