Perplexity: Sonar Reasoning

Perplexity's Sonar Reasoning model, based on DeepSeek R1, exhibits a mixed performance profile. Its speed performance tends to be slower, ranking in the 18th percentile across benchmarks, indicating longer response times compared to many other models. Similarly, its pricing is positioned at premium levels, falling into the 9th percentile for cost competitiveness. However, a significant strength lies in its exceptional reliability, achieving the 95th percentile, meaning it consistently provides usable responses with minimal technical failures. Across benchmark categories, Sonar Reasoning demonstrates strong capabilities in Reasoning (88.0% accuracy, 86th percentile) and General Knowledge (97.0% accuracy, 53rd percentile), suggesting proficiency in complex problem-solving and factual recall. It also performs well in Email Classification (99.0% accuracy, 73rd percentile). Conversely, the model shows notable weaknesses in Instruction Following (34.7% accuracy, 35th percentile) and Ethics (95.0% accuracy, 30th percentile), indicating areas for improvement in adhering to complex directives and nuanced ethical scenarios. Coding performance is average at 81.0% accuracy (51st percentile). Its built-in web search and uncensored nature are additional features.