Exa has released Agent Ultra, the highest effort level of its Exa Agent API, designed for exhaustive research tasks such as large list building and entity enrichment. The service is now live as a hosted API, allowing users to set `effort: "ultra"` to trigger a mode that runs longer than other levels to return more complete results.
- Agent Ultra splits tasks into subtasks and assigns parallel subagents across multiple domains, routing frontier models where needed and faster models elsewhere.
- Exa reports that Ultra outperforms Opus 5.5, GPT-6 Astra, and Perplexity Agent on four benchmarks: WANDR (81.4% soft recall), DeepSearchQA (93.9% F1), WideSearch (58.9% row-level F1), and Company Find-All (2,451 avg passing entities).
- Typical runs complete complex tasks in about 30 minutes, with very hard tasks taking up to 3 hours.
- Pricing is metered at standard Agent rates with a default cap of $20 per run, adjustable via `maxCostDollars` ($1–$100) and `maxDurationSeconds` (300–10,800).
The API targets model providers building training data, financial services conducting diligence, and go-to-market teams enriching account lists, offering a way to exhaustively gather information backed by cited URLs.