Release2026-08-20
Mistral released Agentic Search, a retrieval layer that lets an AI model repeatedly search, open and read inside documents instead of answering from one batch of retrieved text. Mistral reports large accuracy gains on two document question-answering tests over financial filings and scanned government tables, with lower latency and token use. It runs on an organisation's existing search index, in the cloud or on its own servers, via the Mistral Search Toolkit or built into Studio and Vibe.
What changed
Retrieval fetched a fixed set of text chunks and answered in one pass.
What it unlocks
Letting a model open a document, jump to a page or table, and check the figure before answering.
- FinanceBench: 26.7% to 86%
- OfficeQA Pro: 6.3% to 51.9%
- p90 latency 255s to 154s
- token use down up to a third
What you need to act on it
- an existing search index
- Mistral Search Toolkit, Studio or Vibe
- engineering setup for cloud or on-premises deployment
- mistral.ai2026-08-20