Tavily search_depth Basic vs Advanced, Explained

In short: basic search is the lighter option. advanced fetches the result pages and BM25-ranks their content chunks, which can make useful passages easier to find but adds work—and can add latency.

What changes between basic and advanced

Both modes search through SearXNG. The key difference is what Siftdog does with the results afterward.

• Basic: Siftdog uses the search results and blends BM25 relevance with the upstream engine’s order. It skips the advanced page-fetching and chunk-ranking step.
• Advanced: Siftdog fetches each result page, extracts its content, divides that content into chunks, and returns the most relevant chunks using BM25 ranking.

A chunk is a portion of a page’s content, rather than the whole page. This can help when a page covers several topics and only one passage answers your query. The trade-off is that Siftdog must fetch and process pages, so advanced search does more work.

Choose a search depth

Use basic when you want a quicker, lighter search or when the result titles and snippets are enough to decide which pages matter. Use advanced when you want Siftdog to inspect page content and return passages ranked for relevance.

Set the depth in the same /search request:

{
  "query": "how does BM25 rank documents",
  "search_depth": "advanced",
  "chunks_per_source": 2
}

chunks_per_source controls how many relevant chunks Siftdog returns from each source. Keep the number modest if your application has a limited context window; increase it when you need more of each page’s relevant content. The parameter is listed for /search; this example does not assume a default value.

What the benchmark shows

In Siftdog’s 45-query benchmark run on September 25, 2026, advanced mode found answers to all 25 factual questions, compared with 24 in basic mode. Both modes returned official documentation for 9 of 10 technical questions, while advanced ranked an official docs result first in 7 cases versus 3 for basic.

Treat these as results from one small run, not a guarantee. Some upstream search engines were rate-limiting the test machine, and results can change. The benchmark suggests a quality-versus-time trade-off, but your queries and SearXNG setup may behave differently.

Limits

Advanced search depends on Siftdog being able to fetch the result pages. It does not add JavaScript rendering, and it cannot guarantee that a page’s content will be available or answer your query. If you only need to inspect known URLs, use /extract; if you need a broad site traversal, use /crawl or /map instead. See the Siftdog repository for the API details.