Perplexity Agent API effort levels cost less with better quality; ultra added
Perplexity said every Agent API effort level (bundled agent configurations for specific tasks) now costs less at the same or higher research quality, and added an ultra level for the most demanding, long-running requests. It published benchmark changes for medium, xhigh, and ultra.
Note: This article reflects rates and features at the time of announcement, which may have since changed. For the latest pricing, see Subscriptions and API Pricing.
Perplexity Developers announced the update on November 6, 2026 (Singapore time) in an X thread, quoted by CEO Aravind Srinivas. The companion post is Improvements to every Agent API effort level; docs are on Effort.
What changed
- Every Agent API effort level now costs less with the same or higher research quality.
- Effort levels are bundled agent configurations optimized for specific tasks.
- A new ultra level is available for the most demanding, long-running requests.
Benchmarks cited by Perplexity
- medium: score rose from 55% to 77% on BrowseComp and DSQA, with almost 90% lower costs.
- xhigh: held 76.5% on WANDR (Perplexity’s benchmark for long, multi-entity research), with 75% lower costs versus August.
- ultra: highest WANDR score at 85.9%.
How ultra runs
Perplexity says ultra runs three research agents in parallel while a fourth orchestrates, assigns tasks, and merges findings; each child agent has its own research loop and separate context, model, tools, and step budget. Docs list ultra ([email protected]) on anthropic/claude-opus-5-5, with no frozen configuration—call it by name or pin the returned version; background mode is supported.
Available levels
fast, low, medium, high, xhigh, ultra.