Perplexity has updated its Agent API effort levels and added an ultra option for broad, comprehensive research. In its announcement, the company says every effort level now costs less per query while delivering research quality that is higher or unchanged.
Effort levels let developers choose a balance of cost, speed and research depth. Each is a bundled configuration defining the model, step budget, search settings and tools for a particular type of task. Requests that select a level by name, such as effort="fast", automatically receive the updated configuration, Perplexity says.
What the benchmark figures show
Perplexity reports that medium performance improved “from 55% to 77% on BrowseComp and DSQA,” which it describes as public research benchmarks, while costs decreased almost 90%. The announcement does not explain whether that improvement combines results across the two benchmarks or applies separately to each.
For xhigh, the company reports a WANDR score holding at 76.5%, with runs costing about 25% of the August 2026 configuration’s cost. That means about one-quarter of the earlier cost, not a 25% reduction. Perplexity describes WANDR as its own benchmark for long research tasks involving multiple entities.
These are Perplexity’s own performance and cost evaluations. The nearly 90% cost decrease applies to its reported medium comparison, while the xhigh comparison uses the August 2026 configuration as its baseline.
What Perplexity changed
Perplexity for the results: budgets, context handling and extraction lowered run costs, while skills, model selection and parallel agents raised or maintained answer quality.
