The company says the agent decomposes a task into subtasks, researches multiple domains in parallel, and uses a mix of frontier and cost-effective models.
In its launch post, Exa reports results across WANDR, WideSearch, DeepSearchQA, and its internal Find-All Company benchmark. It reports 81.4% soft recall on WANDR at a $18.53 median cost per task, 58.9% on WideSearch at $3.85 per task, and 229 successful rows per task on Find-All Company. The company also notes that some competitor figures were published results while others were run by Exa.