In September 2026, seven AI models were tested on their ability to retrieve current information and verify facts, using the BotHub aggregator for access. The test included models from Perplexity (Sonar Pro, Sonar Reasoning Pro, Sonar Deep Research), as well as ChatGPT, Claude, Gemini, Grok, and DeepSeek.
The test revealed significant differences in accuracy and cost. Simple queries were handled well by most models, but complex tasks, such as creating a table of Central Bank interest rate decisions, led to errors. Costs varied, with a single Sonar response costing between 1.5 and 3.5 RUB, while a Deep Research report cost 98 RUB.
Perplexity's own subscription service does not accept Russian bank cards or the Mir payment system. Users often resort to foreign cards, intermediaries, or pre-registered accounts, though the latter carries risks; in December 2025, Perplexity mass-deactivated accounts linked to promotional codes sold by third parties.
Research from the Tow Center (March 2025) and the European Broadcasting Union (October 2025) highlights the persistent issue of AI hallucinations. The studies found that a significant portion of AI-generated responses contained factual errors or cited non-existent sources, with paid models often providing confident but incorrect information.
