Perplexity Sonar Pro

Perplexity Sonar Pro is the paid tier of the API that Perplexity opened to developers on 21 January 2025, alongside the cheaper Sonar. It is not a frontier model in the usual sense and it is not trying to be one: instead of answering from weights, it runs a web search, reads the results and returns an answer with citations attached. That single design choice is the whole product. It also explains the one number Perplexity actually leads on. On SimpleQA, the standard short-fact factuality benchmark, the company reports an F-score of 0.858 for Sonar Pro against 0.773 for Sonar — ahead of the general-purpose models it was compared with at launch, which have to recall facts rather than look them up. Sonar Pro also returns roughly twice as many citations per search as the base tier and accepts source filtering, which is the feature developers buy it for: an answer you can audit back to a URL. The trade-off is that almost nothing about the model itself is public. Perplexity publishes no parameter count, no architecture and no knowledge cutoff for Sonar Pro — reasonably, since the knowledge cutoff is meant to be "today". Context-window figures of around 200,000 tokens circulate widely but appear only in third-party API listings, not in Perplexity's own documentation, so this catalogue does not treat them as confirmed. Pricing is published and unusually shaped: 3 USD per million input tokens and 15 USD per million output tokens, plus a per-request fee that scales with how much searching you asked for — 6 to 14 USD per 1,000 requests depending on search context size. Base Sonar is 1/1 USD per million with a 5-14 USD per 1,000 request fee. Sonar Deep Research bills differently again, adding citation tokens (2 USD/M), reasoning tokens (3 USD/M) and 5 USD per 1,000 search queries. The important 2026 development is structural. Perplexity's documentation now carries the line "Sonar Chat Completions is now Agent API": the company has moved developers to a new agent-shaped interface that runs a reasoning-acting loop with built-in web search, URL fetching, code sandboxes and MCP servers. Sonar Chat Completions remains supported, but Perplexity itself calls the Agent API more performant and cost-effective for production. Notably, the Agent API also brokers model families from OpenAI, Anthropic, Google, xAI, Z.AI, Moonshot AI and NVIDIA — so Perplexity's developer platform is no longer only a way to buy Perplexity's own model.

▸News

▸Videos

No videos yet.

▸Reviews