In short. Mistral’s web_search accesses a search engine, while web_search_premium also accesses news articles through integrated news provider verification.
Definition: As of 17 October 2026, Mistral documents
web_searchas a tool that accesses a search engine, andweb_search_premiumas a tool that also accesses news articles through integrated news provider verification.
Key facts
- Mistral's Websearch documentation gives both variants search engine access;
web_search_premiumalso accesses news articles through integrated news provider verification. - Mistral's tools overview supports both variants through the Conversations API and Agents API.
- Mistral's Citations & References guide shows how to extract references and resolve their source titles and URLs.
What content can each variant retrieve?
Mistral's Websearch guide describes web search as retrieving recent information or accessing specific websites. The basic variant supplies search engine access. The premium variant combines that access with news articles via integrated news provider verification.
The documented difference concerns the additional news access route. The guide does not specify separate support for PDF files, images, or video for either variant. Its example agent description mentions news, weather, and sports results as subjects for web searches.
How is the tool selected?
Mistral's Websearch agent example enables basic search through the tools parameter during agent creation. This Python configuration fragment matches the example:
tools=[{"type": "web_search"}]
Selecting the documented premium tool type uses this corresponding fragment:
tools=[{"type": "web_search_premium"}]
These fragments belong inside an agent creation request. The example also supplies a model, an agent name, and instructions. After creation, the returned agent identifier is passed as agent_id when starting a conversation. Other tools can be added to the same agent, and the model remains free to search on demand.
How can a search and its sources be checked?
Mistral's output explanation describes tool.execution as the entry recording a search tool's execution. It includes the tool name and start and completion timestamps. The example names web_search.
The answer appears in a message.output entry. Its content contains text interleaved with tool_reference chunks, which identify sources through fields including title, url, and source.
Mistral's separate Citations & References guide shows how an application extracts reference identifiers from response chunks and looks up the corresponding source titles and URLs in the tool result.
Related terms
FAQ
Can I use these tools through Chat Completions?
Mistral's compatibility warning says both variants work with /v1/conversations and the Agents API. Chat Completions does not support these built-in tools because its responses omit the search result references they return.
Where do I find the source URLs?
Read the tool_reference chunks in the answer's content. Mistral's output explanation identifies url as the reference source's URL and title as its title.
Sources
- Mistral AI, "Websearch", undated
- Mistral AI, "Citations & References", undated
- Mistral AI, "Agents Tools Overview", undated
Check how a page reads to answer engines. Open the GEO score analyzer