Exa Debuts Agent Ultra, a Multi-Subagent API for Large-Scale Research and List Creation
Exa has introduced Agent Ultra as the most compute-intensive setting in its Exa Agent API. The hosted system splits research tasks among subagents to handle large-scale list creation and entity enrichment across thousands of sources. According to Exa, it surpasses Opus 5.5, GPT-6 Astra, and Perplexity Agent on four benchmarks, including an 81.4% soft recall score on WANDR.
Exa's Agent Ultra is a hosted tier of the Exa Agent API, chosen by setting effort to "ultra." It breaks a research request into smaller jobs, assigns subagents across domains, and matches frontier or faster models to each step. Exa says typical complex runs take roughly half an hour, while unusually difficult ones may extend to three hours.
The launch targets training-data assembly, financial diligence and KYC work, and go-to-market list building. Ultra can also accept existing rows and omit them from new output. Billing uses the same metered structure as other Agent calls, capped by default at $20 per run, with early finishes costing less. Exa's published comparisons have not yet been checked by outside groups.
Tools like Agent Ultra may change how researchers, analysts, and sales teams gather and verify large datasets, potentially speeding due diligence and market mapping while raising concerns about cost, access, and overreliance on vendor-reported benchmarks. Smaller organizations could benefit from hosted access, but may remain dependent on API pricing and provider controls. If such systems scale, information work could shift toward supervising automated collection and checking citations, though independent evaluation may be needed to assess reliability.