Exa launches Agent Ultra for exhaustive research and list building
Exa has released Agent Ultra, the highest-effort mode of its Exa Agent API. The hosted service divides assignments into subtasks, sends subagents to research multiple domains in parallel, and selects frontier or faster models for different steps. It is intended for large-scale list building, entity enrichment and questions requiring thousands of sources. Ultra is available through the API with effort set to "ultra"; it is not open weight and cannot be self-hosted.
Exa says Agent Ultra achieved the highest results among the compared systems on four benchmarks: WANDR, DeepSearchQA, WideSearch and Company Find-All. Reported scores include 81.4% on WANDR, 93.9% on DeepSearchQA, 58.9% on WideSearch and an average of 2,451 passing entities per Company Find-All task. The comparisons included Opus 5.5, GPT-6 Astra and Perplexity Agent, all at maximum effort. Exa notes that the results are vendor-reported and have not been independently reproduced.
Typical complex runs take about 30 minutes, while especially difficult tasks may last up to three hours. Usage is metered at standard Agent rates, with a default maximum of $20 per run and an adjustable budget from $1 to $100. The API also supports schemas, streaming, time limits and early stopping.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.