--- format: "aidr-story-markdown/v1" id: "186c930fcfd2018322b9e37512a407beedc8c000c084446678ee1956c85a9327" canonical_url: "https://aidr.today/186c930f?lang=en" title: "Exa Launches Agent Ultra: A Subagent Swarm Deep Research API Built for Exhaustive List Building" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-09-26T08:04:08.000Z" category: "Agents" topics: ["agent"] source_urls: ["https://www.marktechpost.com/2026/09/26/exa-launches-agent-ultra-a-subagent-swarm-deep-research-api-built-for-exhaustive-list-building/"] summary: "Exa has released Agent Ultra, the highest effort mode of its Exa Agent API. It coordinates subagents across thousands of sources for list building and entity enrichment. Exa reports it beats Opus 5.5, GPT-6 Astra, and Perplexity Agent on 4 benchmarks, including 81.4% soft recall on WANDR. The post Exa Launches Agent Ultra: A Subagent Swarm Deep Research API Built for Exhaustive List Building appeared first on MarkTechPost ." --- # Exa Launches Agent Ultra: A Subagent Swarm Deep Research API Built for Exhaustive List Building > [Open the canonical story]() **Published:** 2026-09-26T08:04:08.000Z **Category:** Agents **Topics:** agent ## Summary Exa has released Agent Ultra, the highest effort mode of its Exa Agent API\. It coordinates subagents across thousands of sources for list building and entity enrichment\. Exa reports it beats Opus 5\.5, GPT\-6 Astra, and Perplexity Agent on 4 benchmarks, including 81\.4% soft recall on WANDR\. The post Exa Launches Agent Ultra: A Subagent Swarm Deep Research API Built for Exhaustive List Building appeared first on MarkTechPost \. ## Sources - [Story source]()