Introducing Exa Research
Agentic search that doesn't stop until it finds what you need
Agentic search that doesn't stop until it finds what you need
WRITTEN BY
PUBLISHED ON
Jun 4, 2025
Today we are launching /research - agentic search that doesn't stop until it finds what you need. It's particularly useful for automating web research that requires making many searches, then returning insights as structured outputs.
Exa Research is the frontier performance Research API, at the lowest price over common research tasks.
Our research endpoint (pro version) achieves a 94.9% on SimpleQA.
To run this benchmark, we followed the standard SimpleQA protocol, by passing the input query as instructions to the SimpleQA research task and using an LLM as judge. For Gemini, GPT4.1, and Perplexity, we took the highest published results we found.
The endpoint was designed to perform particularly well on structured output. We aggressively parallelize research tasks across fields in the output to help bring overall latency down. The emphasis on structured output also means predictable responses which are easy to integrate into downstream apps.
exa-research is the base "model". It is maximally economical because it adapts model used and webpages read to the difficulty of the task. It's priced at $5 /1k searches and $5 /1k webpages read.
exa-research-pro uses smarter models for every task it is given, maximizing quality. It's priced at $5 /1k agent searches and $10 /1k webpages read.
Reasoning models are used in both models for research planning (more sparingly for exa-research), and are priced at $5/1M reasoning tokens.
The integration should be a breeze - Research is both OpenAI compatible, and available as an asynchronous-first endpoint.
Try it out in the dashboard today! Let us know what you think.
SEE MORE
