Mistral Launches Agentic Search for Complex Enterprise Documents

Mistral Agentic Search is a retrieval layer designed to enable AI systems to navigate, read, and verify information inside complex documents. [1] It introduces a multi-step retrieval loop for finding, inspecting, and verifying information across various data sources. [1] The system builds on existing search indexes by utilizing five specific tools: search, open, navigate, read, and grep. [1] Agentic Search is available through the Mistral Search Toolkit and is built into Libraries in both Studio and Vibe. [1]

According to Mistral, Agentic Search delivers more accurate search results while reducing turns, token use, and latency. [1] On the FinanceBench benchmark, correctness on financial filings increased from 26.7% to 86%. [1] For table-heavy, multi-document questions on the OfficeQA Pro benchmark, Mistral measured a gain of 45.6 points, moving from 6.3% to 51.9%. [1] Furthermore, targeted navigation can reduce p90 latency by up to 39.6%, and token consumption is reduced by up to one-third due to fewer repeated searches. [1]

Mistral's Agentic Search represents an effort to improve how models handle dense, proprietary enterprise data, an area where traditional RAG setups often struggle. By introducing a multi-step loop with explicit tools like 'grep' and 'navigate', the approach seems focused on targeted data extraction rather than simple semantic similarity. The reported improvements on FinanceBench (jumping to 86% correctness) and OfficeQA Pro indicate that this method is particularly effective for complex financial documents and tabular data. What the sources don't address: How Agentic Search performs outside of specific, highly structured benchmarks like FinanceBench.