Exa vs Tavily

The boundary

After recurring free credits are excluded, Exa is lower for paid standard search with up to ten results through roughly 14,300 requests a month. Tavily is lower from roughly 14,300 through 20,000, Exa is lower again from roughly 20,000 through 31,400, and Tavily retakes the lower line from roughly 31,400 through 300,000. Above roughly 300,000 requests, Exa’s linear request meter is lower than Tavily’s published self-serve path. Cent-level rounding creates a few tiny tie bands at the exact crossings, so the calculator reports the whole-request transition rather than pretending one rounded threshold is universal. (Exa pricing, retrieved 2026-07-26; Tavily credit documentation, retrieved 2026-07-26)

Search depth can reverse that result before volume does. Exa publishes separate standard, Deep Search, and Deep-Reasoning Search request rates. Tavily charges one credit for basic search and two for advanced search. The fair comparison therefore matches standard to basic and deeper retrieval to the closest eligible deeper mode. Result count also matters because Exa includes up to ten results in its base search price and adds a meter above that count. (Exa pricing, retrieved 2026-07-26; Tavily credit documentation, retrieved 2026-07-26)

How the billing models differ

Exa is pay-as-you-go by request for Search, Deep Search, and Deep-Reasoning Search, with separate page-content, extra-result, and summary meters. Its recurring account credit reduces small monthly bills but does not turn the public API into a stepped subscription ladder. Tavily offers a recurring free credit allowance, monthly credit bundles, and a pay-as-you-go credit path. Basic and advanced search consume different credit counts, while Extract, Map, Crawl, and Research use their own rules. (Exa pricing, retrieved 2026-07-26; Tavily credit documentation, retrieved 2026-07-26)

Neither reviewed search-pricing page explicitly states whether a failed Search request consumes its normal unit. Tavily does explicitly exempt failed Extract and Map calls, but that statement is not extended to Search in the reviewed credit guide. The baseline therefore uses requested searches at full count and does not manufacture a vendor-specific success rate. (Exa pricing, retrieved 2026-07-26; Tavily credit documentation, retrieved 2026-07-26)

What each offers that the other does not

Exa publishes Contents with full page text and highlights, configurable live-crawl policies, AI page summaries, Deep-Reasoning Search, Monitors, and an asynchronous Agent product. Tavily publishes Extract, Map, Crawl, and Research in the same credit system, with Crawl explicitly combining mapping and extraction. Exa is the direct fit when search plus selected page contents, summaries, or event monitoring is the operation. Tavily is the direct fit when the application must map a site, extract known URLs, or run a combined crawl under one API account. (Exa pricing, retrieved 2026-07-26; Tavily credit documentation, retrieved 2026-07-26)

Who should pick which

Pick by the volume segment only after matching depth and result count. Exa fits lower standard-search volume, the plan-step interval from roughly twenty thousand through 31,400 requests, and sustained volume above 300,000. Tavily fits the two intervening bundle ranges and workloads where Extract, Map, Crawl, or Research replaces separate tooling. Recalculate when returning more than ten Exa results, requesting summaries or contents, or using Tavily advanced search because those operations change the meter. (Exa pricing, retrieved 2026-07-26; Tavily credit documentation, retrieved 2026-07-26)

Search failure billing is not published clearly enough for either vendor, and Tavily’s reviewed docs do not state whether unused paid credits roll over. Those unknowns remain visible rather than being converted into assumptions. (Exa pricing, retrieved 2026-07-26; Tavily credit documentation, retrieved 2026-07-26)