Sonar Pro

Perplexity launched the Sonar API on January 21, 2025, as a real-time generative search service available in two tiers. Sonar Pro is positioned for complex, multi-step queries and offers a 200,000-token context window, compared with 127,000 tokens in the base Sonar tier, along with double the citation depth. Developers can tune outputs via parameters such as Top P and presence penalty. The launch positions Perplexity as a direct competitor to Google and OpenAI in enterprise AI search. Pricing for Sonar Pro is set at $3 per 750,000 input words and $15 per 750,000 output words, with the base Sonar priced at $5 per 1,000 searches plus $1 per 750,000 words. Both tiers emphasize real-time web connectivity for cited, authoritative answers. Perplexity announced the service on its official X account, framing Sonar as the most affordable search API product on the market. The launch accompanies a Pro version with deeper research and retrieval functionality.

Perplexity launched its enterprise Sonar API on January 22, 2025, offering both a basic Sonar tier and a premium Sonar Pro tier for integrating real-time generative AI search into third-party applications. Sonar Pro is priced at $5 per 1,000 searches plus $3 per 750,000 input words and $15 per 750,000 output words, targeting richer answers with additional citations. Perplexity claims Sonar Pro outperforms Google, OpenAI, and Anthropic models on the SimpleQA benchmark. Zoom has already integrated Sonar's real-time web search into its AI assistant, letting users query the web without leaving video calls. Both Sonar tiers support customizable information sources and live internet connectivity for cited answers, differentiating them from training-data-only models. At launch Perplexity reported annual recurring revenue between $5 million and $10 million, a $520 million valuation, and a $73.6 million funding round completed earlier that month.

A third-party benchmark aggregator lists Perplexity Sonar Pro at $3 per million input tokens and $15 per million output tokens as of retrieval around August 2026. On the DeepResearch Bench, Sonar Pro scored 36.2% overall, landing at the 18th percentile among tracked models, with subscores of 33.9% comprehensiveness, 29.7% insight, 43.4% instruction following, 41.1% readability, and 79.7% citation accuracy, though only 16.8% effective citations. The same aggregator places Sonar Pro at the 58th percentile across two medical ethics evaluations auditing value pluralism in clinical AI advice. Results are sourced from third-party benchmark suites rather than Perplexity directly. The page enables side-by-side comparison against multiple other provider models on shared tasks, offering independent technical context for Sonar Pro's current standing.

According to a specifications page citing Perplexity's official API resources, Sonar Pro uses proprietary weights accessed via API under the model ID sonar-pro, with a 200,000-token context window and no dedicated reasoning mode. The model was released in January 2025 and is positioned as a hosted search-and-answer service. Perplexity additionally charges $6, $10, or $14 per 1,000 requests depending on low, medium, or high context usage. Vendor-reported benchmark results include a SimpleQA F-score of 85.8% and a BrowseComp accuracy of 7.33% from Perplexity's own comparison between Sonar Chat Completions and the Agent API. The page notes that Perplexity ended the Sonar Chat Completions endpoint on September 27, 2026, routing synchronous and streaming calls through the Agent API while async Sonar requests were retired. Methodology disclosure flags that the vendor did not specify exact snapshot, search mode, prompts, or evaluator beyond the named SimpleQA F-score.