Perplexity Decisions API adds web grounding: search or fetch a page before scoring
Perplexity’s official changelog (page updated ~2026-11-04 21:44 UTC) adds mode=web_grounding on the Decisions API: Fast Search per question before answering, or fetch a page URL as state first. Each successful search or fetch costs $0.001 on top of input-token billing.
Verified against Perplexity’s Changelog (page dateModified ≈ 2026-11-04T21:44:17Z), Decisions API docs, pricing, and same-day notes from CEO Aravind Srinivas and Perplexity Developers:
Web grounding
Set mode to "web_grounding" on POST /v1/decisions. Before the model answers, the API runs one Fast Search per question (up to 100) and adds results to state as evidence; search_results lists them. Requests without mode are unchanged. Official note: the Python library and TypeScript SDK do not expose mode yet—send it in the request body.
Fetch a page from a URL
With grounding on, pass a single page URL as state. The API fetches the page text, replaces the URL, then searches as usual; the page appears in fetched_results and search hits in search_results. max_staleness (seconds) caps how old a cached crawl may be before a fresh crawl.
Pricing (official)
- Decisions models bill input tokens: $0.02 per 1M for
pplx-decider-v1.2, $0.01 per 1M forpplx-decider-flash-v1.2; output tokens free; no per-request fee. - With grounding, each web search adds $0.001; each successful URL fetch adds $0.001 (failed fetches are not billed). Fetched text and search evidence count as input tokens.
Faithful paraphrase of the official docs and announcement; no extra performance claims.