Exa Launches Agent Ultra Deep Research API with Swarm AI
Exa introduces Agent Ultra API for exhaustive list building, utilizing subagent swarms and extended compute times of up to three hours.
Stock photo for illustration only, not from the actual event
- Exa launches Agent Ultra API designed for exhaustive and deep list building.
- Employs subagent swarms to research multiple domains simultaneously.
- Completes complex tasks in roughly 30 minutes and harder tasks up to 3 hours.
- Available via Exa API under effort: "ultra" with default pricing up to $20 per run.
Exa has officially introduced Agent Ultra, a new subagent swarm deep research API built specifically to handle exhaustive data-collection and list-building tasks. The API is live and accessible on the Exa API platform by configuring the effort parameter to "ultra". Please note that this model does not feature open weights and cannot be self-hosted by users.
The underlying architecture of Exa Agent splits a primary task into multiple subtasks, assigning subagents to research various domains concurrently. It intelligently routes frontier models to steps requiring advanced reasoning while utilizing faster models for less demanding steps. The Ultra mode dedicates the maximum amount of compute resources, running longer than any other effort setting to ensure the most complete results.
Regarding execution duration, Ultra runs typically wrap up complex tasks in approximately 30 minutes, whereas exceptionally difficult assignments can require up to 3 hours to finish. Furthermore, Ultra is capable of expanding existing lists by accepting previously gathered rows and automatically excluding them from newly generated search results.
Deploying a swarm of subagents for deep research represents a strategic shift from traditional single-model prompting, addressing the inherent limitations of LLMs when performing exhaustive data collection. By parallelizing web exploration across distinct domains, developers can harvest more comprehensive datasets, though managing execution duration and API compute costs remains a crucial operational consideration.
Integration is supported through standard Agent run endpoints, accommodating parameters like outputSchema, input.data, and streaming. Developers can implement the API using the Python SDK as demonstrated below:
from exa_py import Exa
exa = Exa()
run = exa.agent.runs.create(
query="Find all companies building browser automation tools in the United States.",
effort="ultra",
)
run = exa.agent.runs.poll_until_finished(run.id, timeout_ms=3 * 60 * 60 * 1000)
print(run.stop_reason)Pricing is metered at standard Agent usage rates, with a default cap of up to $20 per run, and cheaper rates for runs finishing ahead of schedule. Users can customize budgets using maxCostDollars ranging from $1 to $100 and set maxDurationSeconds between 300 to 10,800 seconds. Additionally, OpenAI compatibility is supported on /responses by setting reasoning.effort to "ultra" in streaming or background mode.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment