Perplexity AI Updates: November 6, 2026
Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled
1. Perplexity Launches Teammates, Persistent Agents That Own Ongoing Work Inside Computer, for Max Subscribers
Perplexity launched Teammates, personal agents for long-running work that run inside Perplexity Computer with the user’s tools and files. A teammate coordinates Computer sessions and other teammates on a project, and it remembers project goals, task history and shared files across runs, applying feedback to later tasks (Perplexity’s example is an operations teammate that adds an agency’s two-week notice period to future schedules). Teammates monitor ongoing work and act when things change without waiting for a new query. Colleagues can hand one work or ask for updates after it is added to a Slack channel, and teammates can pass work to each other, for example one finding website bugs and another preparing fixes. They build on Computer’s existing Projects, skills, memory and automations and inherit its tool and Project permissions; each teammate runs in a separate environment with its own browser sessions, saved credentials are encrypted, and users can review completed work, take over the desktop or redirect a teammate mid-task. Setup is done from Teammates in the web app’s left navigation: name the teammate, connect plugins such as Slack, GitHub (with per-repository access) or Gmail, describe the work it should own, where that work comes from and when it should act, and assign it to a project with an initial task. Teammates are available on the web now and on iOS later, for Perplexity Max subscribers only; The Verge puts the plan at $200 per month and reports that each teammate gets its own thread in the app. The blog post is dated November 4, and press coverage reported the launch on November 5. Source The Verge
2. Perplexity Cuts the Cost of Every Agent API Effort Level and Adds Ultra, a Three-Agent Research Mode on Claude Opus 5.5
Perplexity reconfigured the effort levels in its Agent API, the bundled settings (model, step budget, search settings and tools) that callers select with the effort parameter, and says every level now costs less per query with equal or higher research quality; requests that name a level, such as effort="fast", pick up the new configuration automatically, while pinned versions stay fixed. Perplexity reports that medium rose from 55% to 77% on BrowseComp and DSQA while its cost fell by almost 90%, and that xhigh held 76.5% on WANDR, Perplexity’s own benchmark of long multi-entity research tasks, at about 25% of the cost of its August 2026 configuration. The changes include caps on tool calls as well as steps, with reminders as each budget runs down; context compaction that keeps the system prompt, the request and loaded skills; an extraction pass in fetch_url in which a smaller model returns only relevant quoted passages, cutting a 5,561-token page to 293 tokens and input tokens per high query by about 50% (88.0% at $0.073 per query, against 83.5% at $0.094 without it); search results of up to 4,000 tokens instead of 2,000, a new snippet engine and Fast Search at $1.00 per 1,000 requests; skills that load only when needed (preloading every skill raised cost by 13% on BrowseComp and DSQA); and an automated evaluation loop that adopts a new model only if it wins on quality at the same or lower cost. The new ultra level (version ultra@2026-11-05.0, replacing the ultra-preview level listed in October) runs three agents on anthropic/claude-opus-5-5 that each research the whole task, plus a fourth that merges and deduplicates their findings and writes the answer; it scores 85.9% on WANDR at $21.71 per task, the changelog suggests background mode because runs can be long, and ultra has no frozen configuration. Source Changelog
3. Perplexity’s Decisions API Adds URL Fetching With a max_staleness Freshness Control to Web Grounding
Perplexity added URL fetching to the Decisions API’s web grounding mode. This follows the web grounding launch covered on November 5 and adds new parameters, response fields and pricing. When a page URL is passed as state with mode: "web_grounding", the API fetches the page text before searching and puts it in place of the URL. The response returns the page in a new fetched_results field, while search evidence stays in search_results. A new max_staleness parameter sets the maximum acceptable page age, in seconds since the last crawl, and a stored copy older than that triggers a fresh crawl. Each successful URL fetch costs $0.001, on top of the $0.001 per search and the input tokens for added evidence. The changelog also notes that the Perplexity Python library and TypeScript SDK do not yet expose the mode parameter, so callers have to send it in the request body. Source