Perplexity AI Updates: October 31, 2026

1. Perplexity Reworks Agent API Effort Levels, Moves High to GPT-6.1 Terra, and Adds an Ultra-Preview Tier

Perplexity updated the configurations behind its Agent API effort levels. The high level (version high@2026-10-28.0) switches from openai/gpt-6-sol to openai/gpt-6.1-terra, raises max steps from 15 to 40 with a cap of 40 tool calls per run, and lifts web search token limits from 2,000 to 4,000. The xhigh level (xhigh@2026-10-29.0) stays on anthropic/claude-opus-5-5 with openai/gpt-6.1-sol as fallback, raises max steps from 100 to 300, starts context compaction at 150,000 tokens, and no longer sends a system prompt, so frozen configurations no longer send instructions. Earlier October stamps raised low max output tokens from 32,768 to 128,000 and moved fetch_url content extraction to openai/gpt-6-luna across low, medium, and high. A new ultra-preview effort level, called with effort="ultra-preview", targets broad research and data gathering where coverage matters most; its settings are not fully documented, runs can take a long time, and Perplexity suggests background mode. The undocumented ultra name now returns an error, the wide-research alias maps to xhigh, and max_steps accepts up to 500, up from 100. Teams using frozen configurations must copy the updated values to pick up the changes. Source