I benchmarked 4 web search APIs for AI agents. Here's what each one is good at.
Описание и применение
Автор сравнил четыре API веб‑поиска для ИИ‑агентов, выявив, что You.com быстрее и свежей для живых запросов, Parallel обеспечивает наибольшую точность, Exa лучше в точности@5 и страховании, а Tavily превосходит в поиске документации. Выбор зависит от конкретного случая использования.
Задача
Choosing a web search API for AI agents is hard because vendor benchmarks are biased and public tests use generic trivia that does not match agent queries.
Результат
You.com is fastest and freshest for live turn answers; Parallel has highest accuracy overall and in finance; Exa delivers best precision@5 and insurance accuracy; Tavily excels at documentation lookups. No single winner; choice depends on agent use case.
Рабочий процесс
- Define 30 queries with upstream tasks, search strings, parameters and expected answers.
- Send identical requests to each API (10 results, full content, same freshness window and domain filters).
- Run each query 3 times via promptfoo (90 calls per API).
- Record latency, responses and cost.
- Grade answers using SimpleQA rubric and FreshQA strict rules.
- Aggregate metrics: accuracy, precision@5, freshness, latency p50/p95, cost per correct answer.
- Analyze results per use case and overall.
- Publish results, harness, answer keys and dataset for reuse.
Доказательства
I built a small, open web search API evaluation around agent-style calls, and tested You.com, Tavily, Exa and Parallel live.
30 queries, written the way agents call search.
You.com was the only API to surface the day's new Apache Iggy activity, and it was right on that week's funding rounds and agent-memory posts.
Exa was the only API to get both the live wind speed and both county warnings
Parallel edged ahead because it was the only API to find the right week's jobless-claims report.
Tavily was fastest (25 seconds, 3 steps). Every provider's first API call worked on the first try.
- Дата записи: 2026-10-27T19:11:42.944Z