Understand Perplexity before you adopt it.
Perplexity is an AI answer engine, and in its own words it “orchestrates the best AI model for each task” rather than shipping a flagship model of its own; the models a user can select are named as GPT-5.6 Terra, Gemini 3.8 Flash, Claude Sonnet 5, Kimi K3, GLM 5.3 and Grok 4.6. Its developer surface is mid-change, and Perplexity’s migration documentation states it verbatim: “Sonar Chat Completions is now Agent API. Sonar will be supported until September 27, 2026.” Perplexity publishes no version number for the product, so this page records none.
Plan prices and the unlocked-model list on this page come from Perplexity’s pricing page and its Pro page; token and request rates come from its API pricing documentation; and the retirement wording comes from its Sonar migration page and its Agent API blog post. Commercial-use, citation and ownership terms come from Perplexity’s consumer terms of service and its API terms of service, which are two separate documents.
How Perplexity describes the answer engine
Perplexity sells one answer engine rather than a versioned model of its own, and its documentation is organised around two things a reader has to keep apart: the consumer product and the developer API. One entry in that documentation carries a date, and it is the entry that matters most to anyone already building on the API.
Sonar retires on September 27, 2026
Perplexity’s migration documentation carries the notice verbatim: “Sonar Chat Completions is now Agent API. Sonar will be supported until September 27, 2026.” The vendor’s own blog post states the same date in its own words: “On September 27, 2026, the Agent API becomes the main surface for this work, and Sonar tiers retire on the same date.” Perplexity’s documentation was still serving the four Sonar tiers — Sonar, Sonar Pro, Sonar Reasoning Pro and Sonar Deep Research — on the date this page was checked.
The replacement is the Agent API, and the tiers map to presets
Perplexity states the substitution explicitly: “Sonar tiers map directly to presets: sonar to fast, sonar-pro to low, sonar-reasoning-pro to medium, and sonar-deep-research to high.” The presets it publishes are fast, low, medium, high, xhigh and wide-research, so a migration is a rename plus a choice about how much research effort a request should buy.
The engine routes; it is not described as a model
Perplexity advertises no flagship model of its own. Its own wording is that it “orchestrates the best AI model for each task”, and the models a user can select are listed as GPT-5.6 Terra, Gemini 3.8 Flash, Claude Sonnet 5, Kimi K3, GLM 5.3 and Grok 4.6. Reading the product as a single model misunderstands what the vendor says it sells.
Consumer rights and API rights are two different documents
The consumer terms permit personal, non-commercial use only, and they require that published output cite the service. The API terms take the opposite line on ownership — Perplexity states that the customer owns all Output — and no citation clause was found in them. The two documents are separate and do not make the same promises, so which one applies depends on which product you are using.
Two price surfaces, published separately
Consumer plans are listed as Free “$0/month”, Pro “$20/month” and Max “$200/month”, with Pro at “$17/month when billed annually” and Max at “$167/month when billed annually”; Perplexity publishes no full annual total for either. API usage is priced per 1M tokens, with sonar at $1 input and $1 output and sonar-pro at $3 input and $15 output, plus a per-request fee that depends on the request tier.
What Perplexity’s own pages document
Perplexity’s pages document access surfaces, pricing, a retirement date and the rights attached to output. They publish no benchmark scores and no quality claims that can be checked, so this section carries none; everything below is stated on a Perplexity-owned page and is limited to what those pages say.
A model picker rather than a flagship
Perplexity describes the engine as routing, not as a model: “Perplexity orchestrates the best AI model for each task.” Six third-party models are named as selectable — GPT-5.6 Terra, Gemini 3.8 Flash, Claude Sonnet 5, Kimi K3, GLM 5.3 and Grok 4.6 — with the Pro plan listed as unlocking “Access to top AI models” and the ability to “Select between AI models”.
Sonar tiers, with an end date
Sonar, Sonar Pro, Sonar Reasoning Pro and Sonar Deep Research are still documented and were live on the date checked. The same documentation carries the notice verbatim: “Sonar Chat Completions is now Agent API. Sonar will be supported until September 27, 2026.”
The Agent API and its presets
The Agent API is named as the successor, and the mapping is published rather than left to be guessed: “Sonar tiers map directly to presets: sonar to fast, sonar-pro to low, sonar-reasoning-pro to medium, and sonar-deep-research to high.” The presets listed are fast, low, medium, high, xhigh and wide-research.
Documented access surfaces
The web app at www.perplexity.ai, a mobile app for iOS and Android, the Comet browser — which Perplexity calls “A new browser from Perplexity”, on Mac, Windows, iOS and Android — a Perplexity CLI, the API Platform covering the Agent, Router, Search and Embeddings APIs, an Enterprise tier split into Enterprise Pro and Enterprise Max, and Perplexity Computer.
Documented data handling
For the API, Perplexity states: “We do not retain any query data sent through the API and do not train on any of your data.” For consumers it states the reverse default — “For Free, Perplexity Pro and Perplexity Max users, AI Data Retention is enabled by default” — with a control at Account settings, Preferences, the “AI data retention” toggle. For Enterprise, Perplexity states that “your data is never used for AI training purposes.”
Published pricing on both surfaces
Consumer tiers are priced at $0, $20 and $200 a month, or $17 and $167 a month when billed annually for Pro and Max. API pricing is published per 1M tokens and per 1,000 requests, and the Search API is priced at $5.00 per 1,000 requests. Each generative model also carries a separate request fee, banded low, medium or high, charged on top of its token rates.
Documented specifications
Every row below is stated on a Perplexity-owned page, and the pages are named in the notice above. Perplexity publishes no version number for the product, so no row claims one.
- Company
- Perplexity
- Product
- Perplexity answer engine
- Version number
- Not published
- Engine
- Routes across third-party models
- Selectable models
- GPT-5.6 Terra · Gemini 3.8 Flash · Claude Sonnet 5 · Kimi K3 · GLM 5.3 · Grok 4.6
- Sonar tiers
- Sonar · Sonar Pro · Sonar Reasoning Pro · Sonar Deep Research
- Sonar retirement
- “Sonar Chat Completions is now Agent API. Sonar will be supported until September 27, 2026.”
- Replacement
- Agent API
- Agent API presets
- fast · low · medium · high · xhigh · wide-research
- Consumer plans, monthly
- Free $0/month · Pro $20/month · Max $200/month
- Consumer plans, annual
- Pro $17/month · Max $167/month when billed annually
- API input / output, per 1M tokens
- sonar $1/$1 · sonar-pro $3/$15 · sonar-reasoning-pro $2/$8 · sonar-deep-research $2/$8
- Search API
- $5.00 per 1,000 requests
- Access surfaces
- Web · iOS and Android · Comet browser · CLI · API Platform · Enterprise
Documented access and where the numbers come from
Perplexity is sold directly, with no reseller layer to point at. Every figure on this page can be checked against one of the four Perplexity-owned pages below.
Perplexity
The answer engine itself, and the entry point to the consumer product. Perplexity publishes no version number on this page or the others checked.
OpenPlans and pricing
The Free, Pro and Max consumer tiers at $0/month, $20/month and $200/month, and the source of the plan prices quoted above.
OpenMigrating from Sonar to the Agent API
Perplexity’s own migration documentation, and the page the retirement wording is quoted from verbatim.
OpenAPI pricing
Per-token and per-request rates for the Sonar tiers and the Search API, published by Perplexity on its documentation site.
OpenPerplexity FAQ
Does Perplexity have a version number?
No version number was found in the product name across the pages checked — the main site, the pricing page, the Pro page and the API platform page. Perplexity does not publish one, so this page records none rather than assigning one.
What happens to the Sonar API?
It retires on September 27, 2026. Perplexity’s migration documentation states it verbatim: “Sonar Chat Completions is now Agent API. Sonar will be supported until September 27, 2026.” The vendor’s blog post adds that “On September 27, 2026, the Agent API becomes the main surface for this work, and Sonar tiers retire on the same date.” The documented mapping onto the Agent API presets is “sonar to fast, sonar-pro to low, sonar-reasoning-pro to medium, and sonar-deep-research to high”, and the tiers were still live in Perplexity’s documentation on the date this page was checked.
Can I use Perplexity output commercially?
It depends which product, because the rights sit in two different documents. The consumer terms are restrictive — “We hereby permit you to use the Services for your personal, non-commercial use only” — and require that published output cite the service: “You may not (i) publish any Output generated by the Services without clearly citing the Services, or (ii) misrepresent the source of any Output or the fact that it was generated by artificial intelligence.” The API terms take the opposite position on ownership: the customer “retains all ownership rights in Input” and “owns all Output”, and no citation clause was found in them.
How much does Perplexity cost?
The consumer plans are Free at “$0/month”, Pro at “$20/month” and Max at “$200/month”. Billed annually the equivalents Perplexity publishes are Pro at “$17/month” and Max at “$167/month”; the full annual totals are not published. On the API side, rates are per 1M tokens: sonar $1 input and $1 output, sonar-pro $3 and $15, sonar-reasoning-pro $2 and $8, and sonar-deep-research $2 and $8 plus $2 per 1M citation tokens and $3 per 1M reasoning tokens, with the Search API at $5.00 per 1,000 requests.
Does Perplexity train on my data?
The answer differs by surface. For the API, Perplexity states: “We do not retain any query data sent through the API and do not train on any of your data.” For consumers the default runs the other way: “For Free, Perplexity Pro and Perplexity Max users, AI Data Retention is enabled by default”, changeable at Account settings, Preferences, the “AI data retention” toggle. For Enterprise, Perplexity states that “your data is never used for AI training purposes.”
Is Perplexity a model I can pick or self-host?
Perplexity does not describe the answer engine as a model. Its own wording is that it “orchestrates the best AI model for each task”, and the models it lists as selectable belong to OpenAI, Google, Anthropic, Moonshot AI, Z.ai and xAI. Nothing on the pages checked offers weights to download or a self-hosted deployment.