Sonar Reasoning Pro is a specialized web-grounded research API rather than a broadly evidenced frontier generalist. Perplexity’s documentation supports API availability and its multi-step research positioning; Artificial Analysis lists a 127k-token context window and an estimated Intelligence Index of 12. This supports a technical score above narrow models such as Bria 3.2, but below Gemini 3.5 Flash, Kimi K2.5, GPT-5.5, and Claude Opus 4.8, which have substantially stronger broad capability and coding evidence. Its strongest independent signal is search-enabled usefulness: Search Arena placed Perplexity-Sonar-Reasoning-Pro at or near the top alongside Gemini 2.5 Pro Grounding, and the accompanying paper reports rank 1 / 1150.1 for the high variant. That should not be generalized into overall reasoning leadership: the same paper reports only a 28.3% ArenaHard-v2 win rate. No listed DeepSWE or LiveCodeBench result exists, and the supplied SWE-bench and Terminal-Bench/Aider references provide no model result, keeping coding and agentic scoring conservative. The documented API and quickstart make developer onboarding credible. Pricing is materially unclear: official supplied material provides no price, while Artificial Analysis displays $0.00 token fields that should not be treated as a usable pricing commitment. Availability and multimodal support are documented only at a limited level. Public preference is meaningful for grounded search, but evidence quality is moderated by absent reproducible coding benchmarks and variant-specific Search Arena reporting.
