{"id":6007,"date":"2025-12-07T13:46:20","date_gmt":"2025-12-07T17:46:20","guid":{"rendered":"https:\/\/wp.glbgpt.com\/?p=6007"},"modified":"2026-10-11T11:23:39","modified_gmt":"2026-10-11T15:23:39","slug":"what-llm-does-perplexity-use","status":"publish","type":"post","link":"https:\/\/wp.glbgpt.com\/hub\/what-llm-does-perplexity-use","title":{"rendered":"What LLM Does Perplexity Use? Full 2026 Model Breakdown"},"content":{"rendered":"\n

Perplexity uses its own Sonar models alongside models from OpenAI, Anthropic, Google, xAI, Moonshot AI, Z.ai, and NVIDIA.<\/strong> Its published Search lineup includes Sonar 2, GPT-5.6 Terra, Claude Sonnet 5, Gemini 3.7 Flash, Grok 4.6, and Kimi K3; personal Max adds GPT-5.6 Sol and Claude Opus 5. The official Search model table<\/a> is dated September 4, 2026; the separate API catalogue, accessed October 11, also lists newer releases such as GPT-6.1 Sol and Claude Sonnet 5.5.<\/p>\n\n\n\n

The important buying question is which model you can actually select for your task. Compare the Search and API lists below, or use GlobalGPT\u2019s model catalogue<\/a> to explore a separate workspace for research, writing, and coding.<\/p>\n\n\n\n

\"\"<\/a><\/figure>\n\n\n\n
\n
Try Perplexity Now ><\/strong><\/a><\/div>\n<\/div>\n\n\n\n

On this page:<\/strong> Search models by plan<\/a> \u00b7 Sonar explained<\/a> \u00b7 Best, Thinking, and Research<\/a> \u00b7 API models<\/a> \u00b7 Which to choose<\/a> \u00b7 FAQ<\/a><\/p>\n\n\n\n

Which LLMs Can You Choose in Perplexity Search?<\/h2>\n\n\n\n

Perplexity\u2019s advanced-model availability page<\/a> separates the personal Pro and Max plans from Enterprise plans. The following is its September 4, 2026 Search snapshot, accessed October 11, 2026. These are published plan entitlements; trials, staged rollouts, regional requirements, and organization settings can change an individual account\u2019s selector.<\/p>\n\n\n\n

Search model<\/th>Provider<\/th>Personal Pro<\/th>Personal Max<\/th><\/tr><\/thead>
Sonar 2<\/td>Perplexity<\/td>Yes<\/td>Yes<\/td><\/tr>
GPT-5.6 Terra<\/td>OpenAI<\/td>Yes<\/td>Yes<\/td><\/tr>
GPT-5.6 Sol<\/td>OpenAI<\/td>No<\/td>Yes<\/td><\/tr>
Claude Sonnet 5<\/td>Anthropic<\/td>Yes<\/td>Yes<\/td><\/tr>
Claude Opus 5<\/td>Anthropic<\/td>No<\/td>Yes<\/td><\/tr>
Gemini 3.7 Flash<\/td>Google<\/td>Yes<\/td>Yes<\/td><\/tr>
Grok 4.6<\/td>xAI<\/td>Yes<\/td>Yes<\/td><\/tr>
Kimi K3<\/td>Moonshot AI<\/td>Yes<\/td>Yes<\/td><\/tr>
GLM 5.3<\/td>Z.ai<\/td>Yes<\/td>Yes<\/td><\/tr>
Nemotron 3 Ultra<\/td>NVIDIA<\/td>Yes<\/td>Yes<\/td><\/tr><\/tbody><\/table>
Published Search snapshot: September 4, 2026. \u201cYes\u201d means listed access, not unlimited usage.<\/figcaption><\/figure>\n\n\n\n

For everyday paid Search, Pro already covers a broad selection. The model-access reason to consider personal Max is specific: GPT-5.6 Sol and Claude Opus 5 are Max-only in this table. Compare the wider Perplexity subscription benefits<\/a> if you also need higher research allowances or other premium features.<\/p>\n\n\n\n

Enterprise Pro and Enterprise Max list all ten models above, including GPT-5.6 Sol and Claude Opus 5. Administrators can restrict providers or individual models, so a shorter company selector may reflect policy rather than a missing subscription. Our Perplexity Enterprise features guide<\/a> covers the team controls behind that distinction.<\/p>\n\n\n\n

What Is Sonar, and Is It Perplexity\u2019s Only Model?<\/h2>\n\n\n\n

Sonar is Perplexity\u2019s own search-oriented model offering. Sonar 2<\/strong> is the name in the published consumer Search table, while the Agent API uses the identifier perplexity\/sonar<\/code>. Those labels belong to different product interfaces; neither proves that every Perplexity response is generated by the same model.<\/p>\n\n\n\n

An answer involves more than text generation. Perplexity retrieves material, passes relevant information into a model, and presents an answer with sources. Selecting Claude or GPT changes the model available to that workflow; it does not turn Perplexity into the full Claude or ChatGPT application. That distinction explains how Perplexity can use OpenAI models without being ChatGPT<\/a>.<\/p>\n\n\n\n

For example, a question about a company\u2019s latest earnings needs the correct filing and reporting period before it needs elegant prose. A model can produce a fluent answer from the wrong quarter\u2019s document. Open the cited report and check the date and figures when the result informs a decision.<\/p>\n\n\n\n

Best, Thinking, Deep Research, and Model Council: What Changes?<\/h2>\n\n\n\n

In Perplexity\u2019s current model-selection explanation<\/a>, Best is the automatic model option inside Search<\/strong>. Thinking is a reasoning setting on supported models. Deep Research and Model Council are different workflows with their own model-selection rules.<\/p>\n\n\n\n

Option<\/th>What it changes<\/th>When to use it<\/th><\/tr><\/thead>
Best in Search<\/td>Perplexity selects a suitable model for the query.<\/td>Everyday questions when you do not need a specific model.<\/td><\/tr>
Manual Search model<\/td>You choose an eligible model from the selector.<\/td>Repeating a task with a known model or comparing outputs.<\/td><\/tr>
Thinking<\/td>Gives a supported model more time to reason; responses may take longer.<\/td>Multi-step analysis, planning, or coding questions.<\/td><\/tr>
Deep Research<\/td>Perplexity chooses the models automatically for an in-depth report.<\/td>A sourced report that needs more investigation than one search.<\/td><\/tr>
Model Council<\/td>Runs several models on the same question and synthesizes their answers; available on web for Max and Enterprise Max.<\/td>Decisions where comparing model perspectives is useful.<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

Ordinary Search should not be described as running every listed LLM simultaneously. Model Council is the explicit multi-model workflow.<\/strong> Likewise, a news question does not prove that Best selected Grok, and an uploaded image does not prove that it selected Gemini. The task\u2019s subject alone cannot identify the model.<\/p>\n\n\n\n

Start with the result you need: a brief answer, a carefully reasoned explanation, or a longer report. The examples in what Perplexity AI is good for<\/a> can help you choose a workflow before choosing a model.<\/p>\n\n\n\n

Which Models Does the Perplexity API Support?<\/h2>\n\n\n\n

As of October 11, 2026, the Perplexity Agent API catalogue<\/a> lists models beyond the consumer Search snapshot above. Examples include GPT-6.1 Sol, Claude Opus 5.5, Claude Sonnet 5.5, Gemini 3.8 Flash, and Grok 4.7. An API listing does not add that model to a Pro or Max Search subscription.<\/strong><\/p>\n\n\n\n

Provider<\/th>Selected documented Agent API models<\/th><\/tr><\/thead>
OpenAI<\/td>GPT-6.1 Sol; GPT-6 Sol; GPT-6 Luna; GPT-5.6 Terra\/Sol\/Luna<\/td><\/tr>
Anthropic<\/td>Claude Opus 5.5; Claude Sonnet 5.5; Claude Haiku 5.5; Claude Fable 5.1<\/td><\/tr>
Google<\/td>Gemini 3.8 Flash; Gemini 3.7 Flash; Gemini 3.1 Pro Preview<\/td><\/tr>
xAI<\/td>Grok 4.7; Grok 4.6; Grok 4.5<\/td><\/tr>
Moonshot AI<\/td>Kimi K3 \u2014 perplexity\/kimi-k3<\/td><\/tr>
Z.ai<\/td>GLM 5.3; GLM 5.3 Flash<\/td><\/tr>
NVIDIA<\/td>Nemotron 3 Ultra \u2014 perplexity\/nemotron-3-ultra-550b-a55b<\/td><\/tr>
Perplexity<\/td>Sonar \u2014 perplexity\/sonar<\/td><\/tr><\/tbody><\/table>
Selected entries from the Agent API catalogue, accessed October 11, 2026; this is not the consumer Search model selector.<\/figcaption><\/figure>\n\n\n\n

The Agent API combines a selected model with tools such as web search and URL fetching. Perplexity\u2019s Sonar migration guidance<\/a> describes Agent API presets for tasks ranging from short lookups to extended agent work. A preset can bundle model and tool settings, so the preset name and the underlying model are separate pieces of information.<\/p>\n\n\n\n

What Does Model Access Cost?<\/h2>\n\n\n\n

Consumer subscriptions pay for access inside Perplexity\u2019s app and website. Developer requests are metered separately. Perplexity\u2019s subscription comparison<\/a> identifies the API as a pay-as-you-go product and states that it does not include complimentary API credits.<\/p>\n\n\n\n

Access choice<\/th>What you get<\/th>Main buying consideration<\/th><\/tr><\/thead>
Free<\/td>Best plus a limited number of Pro Searches; no manual advanced Search-model selection.<\/td>Suitable for testing the search workflow before paying.<\/td><\/tr>
Personal Pro<\/td>The Pro models in the Search table.<\/td>Choose it if those specific models and the included workflow meet your needs.<\/td><\/tr>
Personal Max<\/td>The Max Search lineup and premium workflows such as Model Council on web.<\/td>Worth comparing when you need a Max-only model or the added workflow.<\/td><\/tr>
Enterprise<\/td>Organization-managed model access and business controls.<\/td>Evaluate privacy and administration alongside the model list.<\/td><\/tr>
API<\/td>Programmatic model and tool use, billed separately.<\/td>Budget for tokens and tool calls rather than a consumer subscription alone.<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

For developers, these are example default-tier token prices per one million tokens<\/strong> in the Agent API catalogue on October 11, 2026. They exclude tool calls, caching adjustments, and any other applicable charges.<\/p>\n\n\n\n

Agent API model<\/th>Input \/ 1M tokens<\/th>Output \/ 1M tokens<\/th><\/tr><\/thead>
GPT-6.1 Sol \u2014 input \u2264272k tokens<\/td>$2<\/td>$10<\/td><\/tr>
GPT-6.1 Sol \u2014 input >272k tokens<\/td>$4<\/td>$15<\/td><\/tr>
Claude Sonnet 5.5<\/td>$2<\/td>$10<\/td><\/tr>
Claude Opus 5.5<\/td>$4<\/td>$20<\/td><\/tr>
Gemini 3.8 Flash<\/td>$0.75<\/td>$3.75<\/td><\/tr>
Kimi K3<\/td>$3<\/td>$15<\/td><\/tr>
Sonar \u2014 perplexity\/sonar<\/td>$0.25<\/td>$2.50<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

A token price is not the complete cost of a research request. A longer answer, more source material, or repeated tool use can change the bill even when the model stays the same. The Perplexity API cost breakdown<\/a> explains how to compare those components.<\/p>\n\n\n\n

\"Earlier
Earlier API billing interface. The pictured $6 is an example account balance, not a current plan price or free-credit offer.<\/figcaption><\/figure>\n\n\n\n

How to Choose a Model for Your Task<\/h2>\n\n\n\n

Choose a Search model from the plan you actually have, then compare it on a task you repeat. A useful comparison holds the question, source material, and requested format constant. Look for errors, missed requirements, and the amount of editing needed rather than choosing by brand name alone.<\/p>\n\n\n\n