t. toolscout.
ToolScout field notes · Product updates

Perplexity API adds ultra effort level and Decisions API web grounding

Perplexity’s November 2026 changelog describes a new Agent API ultra effort level, Decisions API web grounding and URL fetching, and a compatibility-related change to Analytics API credit-usage responses.

Perplexity’s changelog says the Agent API now accepts `effort="ultra"` for broad research and data-gathering tasks where coverage is the priority. The entry says ultra runs may take a long time and can be run in background mode. It runs on `anthropic/claude-opus-5-5`; the changelog also says there is no frozen configuration for ultra, so callers can use its name or pin the returned version. The Decisions API gained a `web_grounding` mode for `POST /v1/decisions`. According to the changelog, it performs a Fast Search for each question—up to 100 questions—and returns the evidence in `search_results`. Developers can also provide a page URL as state in this mode; the API fetches the page text before searching and returns the page in `fetched_results`. The source lists a $0.001 charge for each search and each successful URL fetch, in addition to applicable input-token charges. It also says `max_staleness` can require a fresh crawl when a stored page copy is older than a specified age. The Analytics API’s credit-usage responses currently omit the Credit Source breakdown for `dataset=credit_usage` on two computer-usage endpoints. Total usage and the Model breakdown remain available, the changelog says. Integrations that assume a fixed set of breakdowns should read the returned category list and should not interpret an omitted breakdown as zero usage.

Product details can change. Check linked official information before making a business decision.

Back to the guide