No one has shipped faster than Perplexity AI in the past month. Yet, when CEO Aravind Srinivas launched the startup, he was met with scepticism and criticism, with some dismissing it as nothing more than an LLM wrapper.
Today, however, from asking complex questions to keeping track of the latest details, Perplexity AI is one of the most useful tools for various purposes.
Balaraman Ravindran, an Indian Institute of Technology Madras (IIT) professor and former mentor to Srinivas during his undergraduate years, told AIM that Perplexity is far more than just a ChatGPT wrapper. “It’s a wrapper around LLMs,” he said, adding that a lot of products should be designed that way because many LLMs have complementary strengths.
“If you are an expert user and know what a particular LLM is good for and what applications it excels at, you should be able to choose. You shouldn’t be stuck with whatever LLM the company built the wrapper around,” he explained.
Perplexity lets users pick from a range of powerful LLMs, including Claude 3.7 Sonnet, GPT-4.5, GPT-4o, and DeepSeek R1 among many other LLMs. Each has its strengths—some are great for coding, while others perform better in reasoning and creative writing.
The company has also built its in-house LLM called Sonar on top of Meta’s open-source Llama 3.3 70B. It is powered by Cerebras Inference, which claims to be the world’s fastest AI inference engine. The model is capable of producing 1,200 tokens per second.
Ravindran noted that this could become a more standard way of designing services with LLMs in the future. “I think it’s a great idea that Aravind had, and the speed at which they’re able to add these features is impressive,” he said.
Even NVIDIA chief Jensen Huang has, on many occasions, praised Perplexity AI. He said it is his preferred tutor and called it a “really helpful” tool. He revealed using it daily to learn about a multitude of subjects, including digital biology.
Perplexity AI in India
Perplexity is coming back to its roots. The company recently announced that it will set up its operations in India and has started hiring for the same as well.
Moreover, Srinivas recently announced that he will provide free access to the Pro version for all students, faculty, and staff at the IIT Madras.
“Thanks to Aravind’s generosity, I don’t have to pay for Perplexity,” Ravindran joked. He further revealed that when Srinivas first built the Twitter search bot, he showed him a demo of it over Google Meet.
Notably, payments provider Paytm has also partnered with AI search engine to integrate its AI-powered search into its app to help users make informed financial decisions.
From a marketing perspective, Srinivas is leaving no stone unturned and is tapping into the pulse of the nation—entertainment and cricket. The company recently held a contest during the ICC Champions Trophy 2025, where participants had to ask questions on the Perplexity app, and the winner could win a prize worth ₹1 crore.
Moreover, active X users are likely familiar with the ‘Ask Perplexity’ feature. Perplexity AI has an account (@perplexity_ai) that users can mention in posts or replies to receive sourced, factual answers—often focused on research and data-driven insights. Like a search engine but with AI synthesis, users can tag @AskPerplexity to get quick, cited responses to their questions.
Perplexity AI is on a Roll
Perplexity’s latest release, the Perplexity Windows app, is now available. The Perplexity API also supports model context protocol (MCP), allowing AIs like Claude to access real-time information.
So in 30 days, Perplexity:
— Shweta (@shweta_ai) March 4, 2025
– Hired Jimmy O. Yang as Chief Security Officer
– Added Gemini 2.0 Flash to their AI models
– Given away $1 million for the Super Bowl
– Launched Deep Research
– Open-sourced R1 1776 (a variant of DeepSeek R1)
– Introduced Comet, a browser for agentic…
Meanwhile, Deutsche Telekom, the parent company of T-Mobile, has partnered with Perplexity AI to create a next-generation AI Phone. Running on a custom Magenta AI operating system, it will feature Perplexity Assistant.
The company also open-sourced R1 1776, a version of the DeepSeek-R1 language model that has been post-trained to eliminate censorship and provide factual responses.
However, Srinivas in a recent podcast, admitted that as a company grows bigger, maintaining the same speed and agility becomes challenging. “It’s beginning to happen already a little bit. We’re not as fast as we used to be.”
“We do have staging, deployment testing, A/B testing—all that stuff’s happening, and that’s naturally slowing us down in getting things out to production widely,” he explained.
Moreover, Srinivas also mentioned the issue of monetisation, explaining that while Perplexity delivers high-quality responses, users still go to Google to fulfil their needs, which means Perplexity does not get the financial benefit.
He explained that he wants to give users an end-to-end experience, where they not only find answers but also complete actions—like making a purchase or booking a flight—within the same platform. “They (users) start with a problem in their mind, seek your help, you give them the answers, and you also help them fulfil the action.”
“What you need to build is this amazing orchestration of small models, knowledge graphs, widgets, LLMs, streaming answers, and multi-step reasoning,” he explained. Today, Perplexity AI’s user base stands at 15 million per month and is expected to grow.
