# OpenDS-World tests web-search agents on 200 tasks

> Researchers introduce a live-web benchmark for deep-search agents. The top model reached 39% accuracy, while human attempts reached 21.1%.

Oossa · 2026-10-23 · https://oossa.com/en/opends-world-tests-web-search-agents-on-200-tasks

Researchers have introduced OpenDS-World, a benchmark that tests whether AI agents can research answers across the live web, using search, browser interactions and varied file formats. The benchmark contains 200 tasks, with human checks to confirm the evidence and that each task has a unique answer.

In the researchers’ evaluation of 12 models under one tool setup, the best reached 39.0% accuracy. Human evaluators got 21.1% across 1,000 attempts. The paper also reports persistent problems with coordinating tools and managing long searches; it does not identify the best model in the supplied abstract.

## The facts

- OpenDS-World contains 200 tasks and evaluates 12 models.
- The paper reports 39.0% accuracy for the top model and 21.1% for human attempts.

## Why it matters

For researchers building web-research agents, OpenDS-World offers a live-web test; its results suggest these tasks remain difficult, but the abstract does not show how the scores translate to everyday use.

## Sources & references

1. [Deep Search in Real-World Web Environments](https://arxiv.org/abs/2610.30181) – arXiv cs.IR, 2026-10-23

Last updated: 2026-10-23