Perplexity drops Deep Research for everyone

Perplexity just rolled out Deep Research, a powerful new research agent that turns hours of deep-dive grunt work into a 2 to 4 minute autonomous report.

Unlike standard search queries that grab a few links and spit out a quick summary, this mode executes dozens of web searches, reads through hundreds of sources, runs code, and reasons through material iteratively before handing you a finished artifact.

Even better, Perplexity is making the feature free for everyone on the web, with rollouts to iOS, Android, and Mac coming soon.

An agent that researches like a human

When tackling a complex topic, a good researcher doesn't just look up one query and stop. You read an article, discover a new term, follow a rabbit hole, and update your hypotheses along the way.

That is essentially what Perplexity's Deep Research is doing under the hood. Equipped with reasoning and coding capabilities, the system refines its search plan as it learns more about the subject.

Once it finishes evaluating the sources, it compiles everything into a comprehensive report. Most runs complete in under 3 minutes.

Users can export the output directly into a PDF or text document, or turn it into a shareable Perplexity Page with a single click.

To back up its performance, Perplexity shared two benchmark results. On Humanity's Last Exam, a grueling benchmark covering over 3,000 expert-level questions across 100-plus subjects, the tool hit 21.1% accuracy. Perplexity says this score tops models like Gemini Thinking, o3-mini, o1, and DeepSeek-R1.

On the SimpleQA factuality benchmark, it recorded a 93.9% accuracy score.

Why it matters

The smartest move here isn't just the raw benchmark scores. It is the pricing strategy.

While rivals treat high-compute multi-step reasoning agents as luxury features reserved for expensive tiers, Perplexity is handing free daily queries to everyone, with higher limits for Pro subscribers. That makes it immediately practical for students, market analysts, and engineers who want quick competitive overviews without paying upfront.

We think this shifts the benchmark for search engines. Finding a blue link or a single-paragraph answer is no longer the bar; users will increasingly expect AI to deliver ready-to-read research dossiers.

The real test will be factual reliability under edge cases. Reading hundreds of sources autonomously is impressive, but hallucinating a synthesized connection inside an authoritative-looking 10-page document is dangerous. Still, for everyday market research, it is well worth testing today.