Google Introduces DeepSearchQA Benchmark as AI Agents Enter a New Era

[madtrader]

To validate the accuracy of Gemini 3 Pro and Deep Research, Google released a new benchmark called DeepSearchQA to test complex, multi-step information retrieval.
The agent also performed strongly on two independent evaluations — Humanity’s Last Exam and BrowserComp — though OpenAI’s ChatGPT 5 Pro slightly outperformed Google on browser-based tasks. Google has open-sourced DeepSearchQA to encourage community testing.

[The_Walking_Dead]

Open-sourcing DeepSearchQA is a strong move by Google.

[Capybara_Capybara]

Benchmarks like this are crucial for real-world AI trust.

[Revenant]

Multi-step retrieval is where agents really get tested.

[Rimon Khan]

Big brands validating crypto payments changes the whole narrative.