Perplexity reveals Deep Research, a free rival to Openai and Google that could change everything
The AI war continues to rage and its belligerents go back to blow. Latest example: Perplexity AI has just presented Deep Research. AfterIts partnership with Bouygues Telecomthe American startup is launching an advanced research tool. The latter is positioned, as its name suggests, by direct competitor of those Openai And Google Gemini. However, unlike these two players who reserve their most efficient features for paying subscribers, Perplexity adopts a free approach here, in order to seduce a wider audience.
A gamble on free
However, this approach has a limit, since users of the free version cannot make more than five requests per day, while pro subscribers benefit from 500 daily interactions. And these five requests do not only concern the Deep Research function, but also apply to questions using models that “reason”, namely Deepseek-R1 and O3-Mini. The fact remains that this remains sufficient for moderate use. This allows at least to get an idea before possibly subscribing to a paid offer. And if you are wondering how Perplexity can take the cost of such generosity, know that its new function is based on Deepseek-R1 and its much lower costs.
In any case, by offering this feature for free, Perplexity seems to bet on a massive adoption to establish its place in the cluttered landscape of artificial intelligence. The service is already accessible online and from mobile applications.
The Deep Research of Perplexity is distinguished on paper by its responsiveness and its ease of use. While a detailed report may require several tens of minutes with Openai, Perplexity promises to generate its analyzes in 5 to 10 minutes, thanks to a process of exploring hundreds of sources.
The tool is based on an iterative and thoughtful mechanism. From a asked question, he follows several successive stages. He begins by analyzing a wide range of online content, from various sites and specialized publications, in order to refine his understanding of the subject. Then, he developed a structured report presenting the essential information and the main conclusions. Finally, the results can be shared via a direct link or downloaded in PDF.
This mode of operation is inspired by that of a human expert expert, while considerably reducing the deadlines. The startup was not, of course, to communicate the now usual results of its new tool in the benchmarks for assessing AI models. According to this data, Deep Research notably records a score of 21.1 % in the Humanity's Last Exam test, ahead of models like Gemini Thinking and Deepseek-R1. On the simpleqa benchmark, it reaches a precision rate of 93.9 %.
For whom?
Still according to Perplexity, the diversity of covered subjects makes Deep Research relevant for various user profiles. It allows you to deepen any subject, analyze economic trends, decrypt marketing strategies or even synthesize recent medical research. Nevertheless, it is necessary for the moment to be careful using these kinds of tools. The few tests that we conducted corroborate those of other online media on the fact that the Deep Research Openai has trouble differentiating the reliable sources of those not verified, while that of Perplexity sometimes quotes obsolete studies.
