**Perplexity Unveils Deep Research, Supported by DeepSeek R1**
Perplexity has introduced a new tool titled **Deep Research**, aimed at facilitating thorough research and analysis by scanning the web and generating detailed reports. This capability is powered by a tailored version of **DeepSeek R1**, an open-source AI model.
### **Comparison of Deep Research with Other AI Tools**
Perplexity’s Deep Research is comparable to features provided by **Google** and **OpenAI**, each of which also offers tools named “Deep Research” for **Gemini** and **ChatGPT**, respectively. In addition, **xAI’s Grok 3** features a similar tool named **Deepsearch**.
Nonetheless, in contrast to Google, OpenAI, and xAI, which utilize proprietary models, Perplexity has chosen an open-source model. CEO **Aravind Srinivas** confirmed on X (previously Twitter) that Perplexity’s version of Deep Research is constructed on a **customized** DeepSeek R1 model, enabling the company to provide the tool at a much lower price—**10 to 100 times cheaper** than its rivals.
### **Cost and Accessibility**
Perplexity is offering Deep Research **for free**, albeit with certain restrictions. **Non-paying users** can obtain a limited number of responses each day, while **Pro subscribers** (who pay **$20 monthly**) enjoy unlimited access. This approach contrasts with Google Gemini, ChatGPT, and Grok 3, which restrict their research tools to **paying users** exclusively.
### **Perplexity’s Open-Source Efforts**
Given that DeepSeek R1 is open-source, developers have the freedom to adapt and personalize it. In response to worries about censorship in the original R1 model—especially concerning critiques of the **Chinese government**—Perplexity has released its own **open-source version** known as **R1 1776**. This iteration has undergone post-training to ensure it provides **uncensored, impartial, and factual information**.
### **Challenges and Accuracy Concerns**
Despite its features, Deep Research has its shortcomings. **Decoder** revealed that the tool **misattributed** the phrase **”stochastic parrots”** to AI researcher **Gary Marcus**, when it was initially introduced by **Emily M. Bender**. Users have also raised concerns that Deep Research occasionally delivers **outdated** or **inaccurate** information, which is troubling given its application in **investment and market analysis**.
Srinivas acknowledged these challenges and noted that Perplexity is making efforts to enhance accuracy, particularly in relation to **financial data**, where correctness is crucial.
### **Ongoing Issues with AI Hallucinations**
These inaccuracies underscore a wider challenge faced by **large language models (LLMs)**—hallucinations, or the production of erroneous information, continue to be a significant hurdle. Although enhancements are underway, this issue **may never be entirely resolved**, indicating users should approach AI-generated research with skepticism.