Perplexity Models Explained: Which One Powers AI Search and What It Means for SEO
Perplexity is the citation-first answer engine, and its own models are the Sonar family: Sonar, Sonar Pro, Sonar Reasoning Pro and Sonar Deep Research. The default search experience runs on Perplexity's own model, which the company trained specifically for that mode, and every answer carries numbered citations to its sources. This page compares the Sonar lineup, the third-party models Perplexity hosts, and what it all means for SEO.
The Current Perplexity Model Lineup
Perplexity splits into two catalogs: its own Sonar family, documented at docs.perplexity.ai, and the third-party frontier models it hosts in the consumer picker, which as of September 2026 include GPT-5.6 Sol and Terra, Claude Sonnet 5 and Opus 5, Gemini 3.7 Flash, Grok 4.6, Kimi K3, GLM 5.3 and Nemotron 3 Ultra, per Perplexity's subscription documentation. Perplexity integrates those third-party models with its own search, citations and safety systems, so this hub focuses on the Sonar family that defines the engine. A "Sonar 2" entry also appears in the consumer picker for paid plans, but Perplexity has published no announcement, specs or documentation page for it.
| Model | Released | Context | Modalities | Where it is used | Best fit |
|---|---|---|---|---|---|
| Sonar | Jan 2025; retrained Feb 2025 | 128K | Text | Default search mode, free and Pro | The volume tier; most answers |
| Sonar Pro | Jan 21, 2025 | 200K | Text | Paid tiers and API depth workloads | Complex multi-source questions |
| Sonar Reasoning Pro | Date not published | 128K | Text | Reasoning-grade searches and API | Multi-step questions and analyses |
| Sonar Deep Research | Feb 14, 2025 (mode) | 128K | Text | Deep Research reports | Exhaustive research; easiest citations |
Release Timeline
Most recent first. Dates are from Perplexity's documentation, changelog and blog, linked in the sources below.
Legacy Sonar Chat Completions API sunsets
The old API gives way to the Agent API, which reached general availability in February 2026; the Sonar models continue on the new API.
sonar-reasoning retired
Removed from the API with Sonar Reasoning Pro documented as its replacement, leaving one reasoning tier in the family.
File attachments arrived for Sonar models
Attachment support landed across the family per the changelog, widening what a Sonar query can include.
Deep Research launched
Exhaustive multi-source research reports became a consumer mode, later exposed through the API as sonar-deep-research.
The new Sonar arrived
Perplexity retrained Sonar specifically to enhance answer factuality and readability for the default search mode.
Sonar and Sonar Pro reached general availability
The Sonar API generation replaced the llama-3.1-sonar line, establishing the current family.
How Model Choice Affects Retrieval, Citation and Freshness
The short answer: every Sonar model grounds every answer in live web retrieval with numbered citations, so the family differs in how much it retrieves and how it thinks, not whether it cites. Perplexity describes the pipeline as real-time search over authoritative sources, summarized with numbered citations linking to the originals.
Depth scales up the family: the default reads a handful of fresh sources fast, Sonar Pro doubles the retrieved results, Reasoning Pro searches iteratively between thinking steps, and Deep Research reads hundreds of sources per report. Freshness is the family constant: Perplexity's well-known recency bias means updated pages enter answers within days, which makes it the fastest feedback loop in AI SEO.
What This Means for SEO
The Perplexity playbook is access, freshness and answer-shape: allow PerplexityBot and Perplexity-User, keep cornerstone content visibly updated, and structure pages so a passage can be lifted straight into a cited answer. The complete method is my Perplexity SEO guide, the commercial version is the Perplexity SEO expert service, and multi-engine programs run under the AI SEO expert program.
Then exploit the transparency. Every answer publishes its sources, so citation mining doubles as competitor research; the workflow behind my Arshay Cooper technical SEO, AEO and GEO case study and the visibility gains in my chiropractic clinic AI visibility case study both started with reading who the engines already trusted and closing the gap.
Every Perplexity Model, Explained
Each Sonar model gets its own page: what it is, its key facts with sources, its retrieval behavior, and the checklist for staying visible when it handles the query.
Frequently Asked Questions
Which model powers Perplexity by default?
What is the difference across the Sonar family?
Does Perplexity also use GPT, Claude and Gemini models?
What is Sonar 2?
Why is Perplexity considered the easiest engine to earn citations from?
Do Perplexity citations drive real traffic?
Anshul Rana
I am an AI SEO, AEO and GEO specialist with 8+ years of experience helping businesses get found on Google and AI platforms like ChatGPT, Gemini, Claude and Perplexity. Top Rated Plus on Upwork with a 100% Job Success Score and 1,000+ websites optimized, working with clients across the US, the UK, Australia and India. I run The Digital Geek for agency-level engagements.
Sources
Every model fact on this page comes from Perplexity's own documentation and announcements, listed below and linked inline where each fact appears.
- Perplexity: models documentation
- Perplexity: Meet the new Sonar
- Perplexity: introducing the Sonar Pro API
- Perplexity: introducing Deep Research
- Perplexity: how Perplexity works
- Perplexity: subscription model lineup
- Perplexity: default model help article
- Perplexity: changelog
- Perplexity: Agent API models
Want Perplexity citing your site, not your competitor's?
The free AI SEO audit shows where your site stands in AI search today, and the service program does the work for you.