Perplexity adds a per-session sandbox meter to the Agent API and widens third-party model resale

Perplexity added a sandbox tool to its Agent API — “an isolated container for executing code during an Agent API request” — priced at $0.03 per session. It is the first Perplexity meter that counts neither tokens nor requests: the docs are explicit that “a session covers up to 20 minutes of active use for billing purposes — this is the billing window, not a runtime cap,” and that SDK search queries issued from inside the sandbox bill separately at $0.005 per request, the same rate as web_search.

The Agent API also widened its third-party resale roster. The prior capture named OpenAI, Anthropic, Google and xAI; the current page adds Z.AI, Moonshot AI and NVIDIA, still “at direct provider rates with no markup.” Perplexity keeps zero margin on the model tokens and monetises the surrounding tool calls instead.

Rate cards elsewhere on the developer platform held steady. Sonar token pricing is unchanged (Sonar $1/$1, Sonar Pro $3/$15, Sonar Reasoning Pro $2/$8, Sonar Deep Research $2/$8 plus citation, reasoning and search-query fees), the Sonar request fee still runs $5/$8/$12 per 1,000 for Sonar and $6/$10/$14 for Sonar Pro and Sonar Reasoning Pro by search-context size, and the Search API remains $5.00 per 1,000 requests with no token cost. The docs pricing page also gained an interactive “Estimate your cost” calculator, with the caveat that it is an “estimate only” and that final cost is metered from each response usage field.

Separately, the consumer pricing page at perplexity.ai/pricing now returns a hard 404 — a change first archived on 2026-07-10 — leaving the Free / Pro / Max / Enterprise plan prices reachable only from inside the product or via second-hand sources.