How they use Tavily

Tako tested Tavily advanced search and Tavily Research Pro in comparative research evaluations. Its benchmark methodology dates these runs to July 21, 2026. Search API tests covered SimpleQA, FRAMES, FinSearchComp and Tako’s VerticalRTK research benchmark. Tako synthesized answers from each provider’s results with an LLM and graded them against reference answers with a separate judge model. Providers used default configurations.

For Research Pro, Tako submitted questions to the research agent, let it search and reason end to end, and measured answer accuracy and completion time. Tako reported Tavily Research Pro accuracy of 40.3% on FinSearchComp T2 and 36.9% on T3, with answer times of 104.5 and 283.3 seconds respectively. Tavily advanced search scored 45.8% on the 131-question VerticalRTK research set. It scored 89.9% on SimpleQA, 45.5% on FRAMES, 48.7% on FinSearchComp T2 and 9.9% on T3.

Separately, Tavily and CopilotKit showcased an open-source Research Canvas combining Tavily search, Tako’s structured-data visualizations, LangChain or LangGraph workflows, and CopilotKit’s interactive frontend.