Perplexity has made its research API dramatically cheaper — and, by its own benchmarks, better at the same time. Unite.AI reported on November 5 that Perplexity announced every Agent API effort level now costs less per query, with higher or unchanged research quality across presets. Medium effort improved from 55% to 77% on BrowseComp and DSQA while its cost fell almost 90%. A new Ultra level joins fast, low, medium, high and xhigh: it assigns the same research problem to three agents in parallel and uses Anthropic's Claude Opus 5.5, with background mode for long runs. Perplexity reports an 85.9% WANDR score at 21.71 dollars per task for Ultra, the highest among configurations tested, while at xhigh the WANDR score held at 76.5% at about 25% of its August 2026 cost.
The price cuts are the headline, but the capability changes matter too. Each search can now return up to 4,000 tokens of results, up from 2,000, and a new snippet engine selects text more closely matched to the query. Fast Search is priced at 1 dollar per 1,000 requests. Together, those changes make the API cheaper per query and more useful per query — the combination that tends to drive adoption among developers building research tools.
The Ultra tier is the ambitious end of the range: three agents working the same problem in parallel, with the strongest model Perplexity offers, at 21.71 dollars per task. That is not cheap in absolute terms, but the WANDR score of 85.9% is the highest Perplexity reports for any configuration. For tasks where the answer has to be right — due diligence, competitive research, technical verification — the premium tier is the product.
Perplexity is not cutting prices in a vacuum. The market for AI research APIs — tools that browse the web, read sources and synthesise answers — has become fiercely competitive, with every provider racing to offer deeper research for less money. A 90% cost cut on Medium effort is the kind of move that forces competitors to respond, and it suggests Perplexity believes volume will more than compensate for the lower price per query.
For the SEO world, the relevance is indirect but real. These APIs power the research tools, monitoring products and AI assistants that increasingly decide which sources get cited. Cheaper, better research means more of it — more agents reading the web, more citations being generated, and more pressure on the pages that want to be cited. The cost of being read by machines is falling; the value of being worth reading is rising.
Every research API is a new reader of your content — and a reader that decides, in milliseconds, whether your page is a source worth citing. Perplexity's improvements mean its agents will read more pages per task and choose snippets more precisely. Pages with clear structure, specific facts and quotable sentences will be picked; pages that bury their point will not.
The practical implication is the same one AI search keeps delivering: write pages that a machine can use. Lead with the answer, support it with specifics, and make the key sentence easy to lift. The agents are getting cheaper and smarter, which means they will be reading more of the web — and the pages they cite are the ones built to be cited.
The cost cuts also change who can afford to build with research APIs. At the old prices, deep research was a premium feature for well-funded products; at the new prices, a small team can run thousands of research queries for the cost of a single analyst's afternoon. That democratisation matters because it multiplies the number of agents reading the web — more products, more queries, more citations being generated by machines rather than by people. For publishers, that means the audience for their content is increasingly synthetic: agents that read, summarise and cite on behalf of humans. Writing for that audience is not a different skill from writing for humans, but it rewards the same qualities more sharply — clarity, specificity, and sentences that survive being lifted out of context.
Sources: Unite.AI, November 5, 2026.
Related reading on this site: ai-visibility-vs-rankings-measure-wrong-thing, google-search-queries-five-year-high, reddit-stock-google-traffic-down-huffman, ai-agents-three-ways-websites-treat-them, google-protect-important-urls-deindexing, and A Connecticut Grocery Chain Outranks Wikipedia on Google for 'Apples'.
Linkbaits helps you create content that earns citations and backlinks. Start with the free tools and build your first linkable asset this week.
Start free trial