Perplexity cuts Agent API effort-level costs and launches Ultra
Perplexity updated every Agent API effort level to run at a lower per-query cost, reporting that Medium rose from 55% to 77% on BrowseComp and DSQA while costing almost 90% less.
AT A GLANCE
Key points
- Medium rose from 55% to 77% on BrowseComp and DSQA while costs fell by almost 90%.
- Xhigh held a 76.5% WANDR score at about one quarter of the August configuration’s cost.
- Ultra merges the outputs of three parallel research agents with a fourth agent.
- Effort levels jointly configure the model, step budget, tool calls, search settings, and skills.
What is changing?
Perplexity updated the Agent API effort levels that balance per-task cost, speed, and research depth. According to the company, every level now runs at a lower per-query cost while quality is higher or unchanged. The Medium configuration rose from 55% to 77% on the combined BrowseComp and DSQA evaluation while costs fell by almost 90%. Xhigh held its WANDR score at 76.5% while running at about one quarter of the cost of its August 2026 configuration.
The new Ultra level sends the same task to three research agents in parallel and uses a fourth agent to merge their findings. Perplexity reports an 85.9% WANDR score at $21.71 per task for Ultra. The update uses step and tool-call budgets, context compaction, page extraction, model selection, and task-specific skills to control cost. Because provider benchmarks do not guarantee results on real workloads, teams should separately measure quality, latency, and total cost on their own query mix.
Explore the original sources
Source date: 6 November 2026.
Published by AI Radar: 6 Nov 2026, 07:05 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…