Perplexity Sonar API

Sonar is Perplexity's family of search-grounded generation models, sold as an API rather than a chat app: a call retrieves live web results and returns an answer with citations, so grounding is the product. The tiers are priced per million tokens — Sonar at $1/$1, Sonar Pro at $3/$15, Sonar Reasoning Pro at $2/$8 chain-of-thought, Sonar Deep Research at $2/$8 plus separate per-thousand charges for citations, reasoning and searches. Around them sits a metered toolbox: the Search API at $5 per 1,000 requests and agent endpoints billed $5–$14 per 1,000 by search-context tier, all pay-as-you-go with no subscription. Perplexity's docs now present the surface as Router, Agent API, Search and Embeddings, with Sonar inside it.

Where it loses: if the question has no freshness requirement, a general model plus your own retrieval costs less per token, and long-document reasoning is not what this line is for. Grounding is also not verification — a citation can point at a thin page and the synthesis can still be wrong, so products quoting Sonar should show the sources, not just the answer. Where it wins is anything that must be current: news, prices, filings, release notes and docs that drift. Before budgeting, check how many searches a request triggers, since retrieval is metered separately from tokens and multi-hop queries multiply the sticker rate, which usage tier your account sits in, and whether you want Sonar end-to-end or the Search API feeding a model you already pay for.