Sonar Reasoning Pro is a reasoning-oriented model that blends Perplexity's search-grounded Sonar family with chain-of-thought capabilities drawn from DeepSeek R1. It is positioned as an advanced analysis model that combines multi-step logical reasoning with enhanced web-grounded information retrieval, making it suitable for tasks that demand both deliberation and up-to-date factual grounding. The model's architecture is intentionally tuned for chain-of-thought processing, allowing it to break complex questions into intermediate reasoning steps before producing a final answer, while leveraging its search backend to pull in supporting evidence along the way. It can accept text and image inputs to enrich queries, though image inputs paired with structured outputs are not supported within its thinking workflow, and its responses are delivered as text. Practical use cases highlighted by the documentation center on complex multi-step analysis, deep research synthesis, and decision support where step-by-step reasoning improves answer quality.

Rather than a from-scratch pre-training effort, Sonar Reasoning Pro is built on the DeepSeek R1 reasoning model and inherits DeepSeek's chain-of-thought lineage, then layered with Perplexity's retrieval and search augmentation so the deliberation can cite fresh, web-grounded evidence. This pairing is the model's defining advancement: reasoning depth from R1 combined with Perplexity's real-time information access, framed as the replacement for the earlier sonar-reasoning model, which was deprecated in late 2025. Comparative analyses describe it as faster and more cost-efficient than related Sonar variants while delivering a higher intelligence benchmark and an explicit reasoning mode, favoring scenarios where thoughtful analysis must be combined with retrieved facts. Forward-looking, it is aimed at developers building research assistants, analytical copilots, and decision-support tools that benefit from transparent multi-step thinking paired with live web context, and the documentation stresses that customer data is not used for training, supporting private and exploratory usage.