Sonar Reasoning Pro is an advanced language model developed by Perplexity, designed to combine deep logical analysis with real-time information retrieval. It serves as a premier tier within the Sonar model family, specifically optimized for tasks requiring multi-step reasoning, complex problem-solving, and the synthesis of information from multiple web sources.
The model is powered by the DeepSeek R1 architecture, utilizing a Chain-of-Thought (CoT) process to break down user queries into smaller, verifiable logical steps. This internal thinking process allows the model to perform iterative searches, evaluate conflicting data, and reduce hallucinations, resulting in highly factual and cited responses. Users can often view the model's reasoning process during the generation phase, providing transparency into how the final answer was derived.
With a context window of 128,000 tokens, Sonar Reasoning Pro is capable of processing extensive documentation and broad search results simultaneously. It is particularly effective for technical research, academic synthesis, and professional decision-making where up-to-date accuracy and logical consistency are critical.