Updates·February 15, 2025, 01:00

Perplexity launches cheaper AI researcher to take on OpenAI

AI-generated from the sources listed, with automated source checks.

Perplexity has launched Deep Research, which searches and writes reports on its own in a few minutes. The price is far below OpenAI's, but the quality figures come from Perplexity itself, which has previously been criticized for misleading comparisons.

The US AI search service Perplexity launched a new tool called Deep Research in mid-February 2025. It is available on the web, and apps for iOS, Android and Mac are on the way.

The idea is that the tool does the research work for the user. According to TechCrunch, it searches repeatedly, reads documents and decides for itself what to do next. The result is a detailed report that can be downloaded as a PDF or shared as a Perplexity Page. According to The Decoder, the tool runs dozens of searches and goes through hundreds of sources.

Faster and cheaper than OpenAI

The sources don't entirely agree on timing: The Decoder reports 1-2 minutes per report, while TechCrunch says under three minutes. According to TechCrunch, OpenAI's comparable tool takes 5-30 minutes.

Price is the big difference. According to The Decoder, Perplexity offers 500 queries a day for $20 a month, while OpenAI charges $200 a month for 100 queries. According to The Decoder, Perplexity CEO Aravind Srinivas has described the price as 10-100 times lower than competitors'.

Free users can also try the tool, but with a limited number of queries per day. The Decoder cites five a day. According to TechCrunch, Pro subscribers get unlimited queries.

According to The Decoder, the tool is built on a customized version of the Chinese company DeepSeek's R1 model, which can "think" in steps before it answers. According to the outlet, that is what keeps costs down. The information comes from The Decoder, and TechCrunch does not mention DeepSeek.

Scores on a hard test

Perplexity has measured the tool on Humanity's Last Exam, a test of expert-level questions. TechCrunch reports that Deep Research scored 21.1%. That is above Gemini Thinking (6.2%), Grok-2 (3.8%) and GPT-4o (3.3%), but below OpenAI's Deep Research (26.6%). The Decoder gives 20.5% for the DeepSeek version, just behind OpenAI. The figures come from Perplexity itself.

The Decoder notes that Perplexity has previously published misleading comparisons on OpenAI's SimpleQA test, in which a service with internet access was measured against models without it.

Still requires critical reading

The Decoder points out that the tool can hide errors in long reports, so they should be checked by a human. TechCrunch also refers to The Economist, which has pointed to limited creativity, reliance on easily accessible sources and a risk of less independent idea development in OpenAI's Deep Research. TechCrunch judges that this probably applies to similar tools such as Perplexity's as well.

For everyday users, the takeaway is that a report from a tool like this is a good starting point, but not a final answer.