Exa Launches Agent Ultra: A Subagent Swarm Deep Research API Built for Exhaustive List Building

Exa has released Agent Ultra, a new API tool designed for large-scale research tasks like building extensive lists and gathering information from thousands of sources. The system works by splitting tasks into subtasks and assigning multiple subagents to research different areas simultaneously, choosing between frontier models and faster models depending on what each step requires. According to Exa's testing, Agent Ultra outperformed competing systems on four research benchmarks, with typical complex tasks completing in about 30 minutes and very difficult tasks taking up to 3 hours. The service is available now as a hosted API with pricing up to $20 per run by default, adjustable between $1 and $100.