Perplexity API Pricing After Sonar Chat Completions Support Ended

Affiliate disclosure: some links in this article are partner links. If you start a paid plan through them, imisofts may earn a commission at no extra cost to you. We only recommend tools we actually use to run client campaigns.

Short answer. Perplexity ended support for Sonar Chat Completions on 27 September 2026, and its pricing page no longer lists a Sonar rate. Synchronous and streaming Sonar calls still work while Perplexity reformulates them as Agent API requests, model by model; asynchronous Sonar calls are no longer supported. On the Agent API you pay each model's per-token rate plus tool fees: $2.50 per 1,000 web search calls, or $1.00 with Fast Search, and $0.50 per 1,000 page-fetch calls. The standalone Search API costs $5 per 1,000 requests, or $1 with Fast Search, with no token charges. Perplexity's own benchmark tables put its cheapest preset, fast, at $0.0014 to $0.0018 per request, against $0.0059 to $0.0070 for Sonar. We earn a commission if you buy Apify or Bright Data through our links; Perplexity pays us nothing.

Every price and limit below was read on 19 October 2026 on Perplexity's, Apify's and Bright Data's own pages; the totals are our arithmetic. We did not run billed tests on any of the three or test answer quality.

What happened to Sonar

Perplexity's migration guide says Sonar Chat Completions support ended on 27 September 2026. Synchronous and streaming requests keep working and are being reformulated as Agent API requests, rolling out gradually by model. Asynchronous Sonar requests are no longer supported and need to move to the Agent API's background mode. The changelog tagged the migration as a deprecation in July 2026.

Sonar the model has not disappeared: perplexity/sonar is listed inside the Agent API at $0.25 per million input tokens and $2.50 per million output tokens, with web search billed separately. What has ended is support for Sonar Chat Completions, and its rate card is off the pricing page. Perplexity suggests these presets as starting points:

Above high sits xhigh, which runs Claude Opus 5.5 for up to 100 steps with standard web search, finance search and a code sandbox.

Two migration details change the bill. Sonar's search_context_size setting has no named equivalent; the Agent API's web search tool takes token budgets instead, and you are charged for the exact number of search context tokens consumed, not the budget you set. And presets are dynamic: Perplexity updates a preset's configuration, which can include its model, as evals show improvements, aiming to stay in the same cost band, unless you copy the preset's current values into a frozen configuration.

Perplexity API prices on 19 October 2026

The pricing page now covers the Search, Agent, Orchestrator, Embeddings and Decisions APIs. From the pricing and Agent API models pages, the lines most relevant to Sonar users:

Priority processing bills twice the listed token price and flex half, on models that support those tiers; GPT-6 Luna and GPT-6 Sol do, Claude Opus 5.5 lists neither. The fast preset runs on priority, so its GPT-6 Luna tokens bill at $0.20 and $1.00 per million.

The Search API bills each successful request, and one request can carry up to five queries for the price of one. It does not bill invalid, rate-limited or upstream-failed requests, but it does bill a successful response that returns no results. It returns 10 ranked results by default, up to 20.

The old Sonar rate card, and what we could not confirm

The pricing page no longer shows Sonar, but a copy of its rate card remains in the data behind the pricing calculator on Perplexity's Agent API Models page:

Sonar's per-request fee was tiered by search context size. We could not confirm whether reformulated Sonar calls now bill on this card or at Agent API rates, or whether Sonar also billed search content as tokens; the Agent API does.

Google's AI Overview for "perplexity api pricing" (US English) told us on 19 October that standard Sonar models carry no per-request search fees and that search and citations are included in the per-token price. Its token rates match the card above, but that card carries request fees of $5 to $14 per 1,000, and the Agent API bills every web search as its own line.

What a request costs on Perplexity's benchmarks, by preset

Perplexity's benchmark page prices each Sonar model and preset per request on its own workloads, with no date, model or preset versions given:

On Perplexity's own workloads, each suggested replacement costs less per request. Fast costs 80% less than Sonar on BrowseComp and 69% less on DeepSearchQA, and 91% less than Sonar Pro on both. Low costs 63% and 54% less than Sonar Reasoning Pro. High costs 22% and 21% of Sonar Deep Research, which bears out Perplexity's "under a quarter", and low costs about 1% and 2% of it, consistent with Perplexity's "1-2%". The exception is xhigh, Perplexity's top-quality tier rather than its Deep Research replacement: it costs 28% more per request than Deep Research on BrowseComp ($0.78 against $0.611) while scoring 85.33% against 29.10%, and 61% less on DeepSearchQA.

Your queries are not Perplexity's benchmark, so check list prices. Perplexity's pricing calculator assumes a fast run reads 1,000 input tokens and writes 500; it labels these median values from representative runs, not billed values, and lists them beside an older model. At the fast preset's priority rate that is $0.0007 in tokens plus $0.001 for one Fast Search: $0.0017 per request, or $1.70 per 1,000, inside the benchmark's range. Without priority it would be $0.00135. The same token counts on the old Sonar card come to $0.0065 at low search context, $0.0095 at medium and $0.0135 at high. That is rate-card arithmetic at identical token counts, not what the same query cost. Inside the Agent API, at the same token counts, perplexity/sonar with one Fast Search comes to $0.0025, or $0.0040 with standard search. Actual cost is metered per response, and a completed response can report it in its usage cost field.

Rate limits, credits and refunds

Perplexity sells API credits in advance. Usage tiers rise with lifetime purchases and never drop: Tier 1 from $50, Tier 2 from $250, Tier 3 from $500, Tier 4 from $1,000 and Tier 5 from $5,000. Agent API limits apply per organization, shared across models and keys: 1, 3, 8, 17, 33 and 33 requests per second from Tier 0 to Tier 5, so the step from $1,000 to $5,000 of purchases adds no Agent API throughput. The Search API allows 50 query units per second on every tier, and a five-query request is one billed request but five rate-limit units.

If credits run out, Perplexity blocks your API keys until you add more; an optional Automatic Top Up refills the balance when it drops below $2. A Perplexity subscription is not required to use the API. We found no free API allowance on the pricing page or in the API billing help article, and the Pro billing FAQ, modified on 17 October, does not mention the API. The help center, modified on 14 September, says refunds "can be processed" if reported within 14 days of purchase and the credits are unused. The API terms, updated on 23 January 2026, still name "Sonar by Perplexity" and say payments are nonrefundable except where the terms expressly say otherwise.

Building your own: search APIs plus a model

Sonar sold search and an answer in one call. The Agent API already lets you pick a model from several providers and freeze its configuration. If you would rather call a model on your own account, you buy search separately and pay that model to read the results. Per billed call at listed paid rates:

At those rates Perplexity's Fast Search is the lowest per call, below both of our affiliates. The units differ, though. A Perplexity request can carry five queries and returns 10 results by default; Bright Data bills per successful request; Apify bills per query and per page fetched, up to three by default, and adds $0.000225 per GB of memory on normal runs. Perplexity also bills successful requests that return nothing, and Apify charges for fetched pages that hold no text. We did not test speed or result quality on any of them.

The comparison that hurts our affiliates most: Perplexity's benchmark lists the fast preset at $0.0014 to $0.0018 per request on its own workloads, $1.40 to $1.80 per 1,000, for a run that returns an answer. Bright Data charges $1.50 per 1,000, or $0.0015 a request, on pay as you go for search results with no answer, so a model adds cost on top. An Apify default query, one search plus up to three pages, costs at most $10.32 per 1,000 on the Free plan through Standby, or $6.02 on the $999-a-month Business plan, before a model reads anything; normal runs add $1.80 per 1,000 at the default 8 GB. Apify's $5 of free usage covers about 484 such queries. Bright Data gives every new account 5,000 free credits a month, one per SERP request, from a pool shared with its other APIs; its menu says "starts from $1/1k req", a rate we could not tie to any plan on its pricing cards.

So the alternatives earn their place for specific jobs. For Google's own results and search verticals such as Shopping, Maps, News or Jobs, use Bright Data's SERP API. For Google's top pages as Markdown, priced the same per page with either of its scraping tools, use Apify's RAG Web Browser. For raw results feeding your own model, Perplexity's Search API with Fast Search is the lowest listed price here, $1 per 1,000 requests. Our AI agent search API comparison prices Tavily, Exa, Brave and others per 1,000 calls, and our Custom Search API alternatives covers raw Google result feeds.

Which Perplexity API route fits which job

• Answers with citations, the old Sonar job: the Agent API's fast preset, or low for light multi-step research. If you freeze fast to pin its model, the changelog says to set the service tier to priority to match it; default processing bills the listed token rate, half the priority rate.
• Deep research reports: high, which Perplexity's benchmark puts at under a quarter of Sonar Deep Research per request. Use xhigh when its higher scores, 85.33% against 78.00% on BrowseComp and 93.00% against 89.60% on DeepSearchQA, are worth about 5.7 and 1.8 times high's cost per request.
• Raw results for your own model or index: the Search API, batching up to five queries per request.
• Google rankings, verticals or page text: Bright Data or Apify, at the prices above.
• Being cited by Perplexity rather than calling it: that is AI search optimization, not an API bill, and you can track Perplexity referrals in GA4.

If you would rather have it built, our AI automation agency can scope the pipeline and cost it before it ships.