Perplexity, an AI search engine startup, introduced that its in-house mannequin, Sonar, might be obtainable to all Professional customers on the platform. Now, customers with the Perplexity Professional plan could make Sonar the default mannequin by way of settings.
Sonar is constructed on high of Meta’s open-source Llama 3.3 70B. It’s powered by Cerebras Inference, which claims to be the world’s quickest AI inference engine. The mannequin is able to producing 1200 tokens per second.
“We optimised Sonar throughout two vital dimensions that strongly correlate with consumer satisfaction – reply factuality and readability,” Perplexity introduced, indicating that Sonar considerably improves the bottom Llama mannequin on these points.
Perplexity revealed that their evaluations discovered that Sonar outperforms OpenAI’s GPT-4o mini and Anthropic’s Claude 3.5 Haiku and affords efficiency parity with the larger fashions GPT-4o and Claude 3.5 Sonnet.
Moreover, Perplexity mentioned Sonar is 10 occasions sooner than Google’s Gemini 2.0 Flash.
Not too long ago, French AI startup Mistral revealed its app, Le Chat, which claimed to be the quickest AI assistant within the competitors. Throughout our testing, we discovered it to be sooner than all different fashions. Gemini 2.0 Flash, alternatively, got here in second. Like Perplexity’s Sonar, Mistral’s Le Chat can be powered by Cerebras Inference.
Not too long ago, Perplexity additionally introduced the provision of the highly effective DeepSeek-R1 mannequin on the platform, hosted on servers in the US.
A couple of weeks in the past, Perplexity introduced that the Sonar API is offered in two variants: the Sonar and the Sonar Professional. The corporate additionally referred to as it probably the most inexpensive API available in the market.
The corporate mentioned Sonar Professional is “ultimate for multi-step duties requiring deep understanding and context retention”. Furthermore, it supplies “in-depth solutions” with twice the citations of Sonar. The Professional model prices $3 per million enter tokens, $15 per million output tokens, and $5 per 1,000 searches, with a number of searches allowed.
The Sonar plan is easier. It fees $1 per million tokens for enter and output and $5 per 1,000 searches, with just one search per request.
The publish Perplexity Launches Sonar for Professional Customers; Efficiency on Par with GPT-4o, Claude 3.5 Sonnet appeared first on Analytics India Journal.
