SUTRA vs Perplexity Models

In this page, we showcase some of the key characteristics of SUTRA models and how they differ from Perplexity AI’s sonar-medium-online models. sonar-medium-online are online models from Perplexity, that just like SUTRA models, leverage information from the internet to augment their responses. In our experimentation, we noticed that at times sonar-medium-online models get factually incorrect information. Similarly, these models have limited multilingual capabilities, especially for Indic and Asian languages.

​

Online Queries

Question: “How was the state of the union address?” [asked on 03/08/2024 8AM PST, the morning after SOTU address was delivered by President Joe Biden]

Response from SUTRA: It understands the question and gets most recent information from internet and formulates and up-to-date response with minimal latency.

President Joe Biden delivered 2024 state of the union
address yesterday. He talked about issues on health care,
climate change, humanitarian support for Gaza and
War in Ukraine and updating tax rates for rich.

Response from sonar-medidum-online: Both SUTRA-ONLINE and sonar-medium-online are able to leverage information from internet and formulate accurate and factually correct response

In his 2024 State of the Union address, President Joe Biden
passionately defended his record, criticized former President
Trump's policies and actions, announced plans for a Gaza pier
to increase humanitarian aid, proposed a minimum tax rate for
billionaires

​

Hallucination

In another query, that required use of up-to-date information, both models had access to internet. However, SUTRA was able to leverage this information precisely providing hallucination-free responses.

Question: “Who won the India vs England cricket match from yesterday??” [asked on 03/08/2024 8AM PST, while 5-day Test match between India and England was on-going] Response from SUTRA-ONLINE: It understands the question and gets most recent information from internet and formulates and up-to-date response with minimal latency.

Yesterday’s India vs England cricket match was the first
day of 5th Test. At stumps India was at 135/1, trailing by
83 runs. So technically, match is still on-going.

Response from sonar-medidum-online: While SUTRA answered it correctly, sonar-medium-online hallucinated with it’s response and provided summary about a match from Oct’2023.

India won the cricket match against England yesterday by 100 runs
in the ICC World Cup 2023. This victory helped India maintain their
unbeaten streak in the tournament.

​

Multilingual Queries

SUTRA models are state-of-the-art multilingual capabilities and can respond in wide ranging languages. Furthermore, they can perform number of tasks in these languages from instruction following to language generation. Following is an illustrative example to show-case this use case.

Question: “Niyutran citaral parriya vilakkattai aintu vakkiyankalil kotunkal” (“Give an explanation of neutron scattering”)

Response from SUTRA: It understands that the question is not just in Tamil, but transliterated Tamil. It understands it and provides response in both Tamil and translated English

நியூட்ரான் சிதறல் என்பது பொருட்களின் அமைப்பு மற்றும் பண்புகளை ஆய்வு செய்ய
பயன்படுத்தப்படும் ஒரு நுட்பமாகும்.
[EN: Neutron scattering is a technique used to study the
structure and properties of materials.]

Response from sonar-medidum-online: It was unable to understand the question and it responds accordingly

I'm sorry, but I'm unable to provide a relevant response to
your query as it appears to be in a language or format that
I do not recognize.