Skip to main content
MindStudio
Pricing
BlogAbout
My Workspace
Text Generation Model

Sonar Deep Research

Perplexity's exhaustive deep research model that autonomously searches hundreds of sources to deliver expert-level analysis and comprehensive reports.

PublisherPerplexity
TypeText
Context Window128,000 tokens
Training DataFebruary 2025
Input$2.00/MTok
Output$8.00/MTok

Autonomous multi-source research with cited reports

Sonar Deep Research is a text generation model developed by Perplexity AI, released in February 2025. It is designed specifically for complex, multi-step research tasks that require gathering and synthesizing information from a large number of web sources. Rather than returning a single retrieved answer, it autonomously plans a research strategy, conducts dozens of iterative web searches, evaluates the results, and refines its approach before producing a detailed, citation-backed report. It operates with a 128,000-token context window, allowing it to handle substantial volumes of text and references within a single session.

Sonar Deep Research is best suited for tasks where thoroughness and accuracy take priority over response speed, such as academic research, market analysis, competitive intelligence, and due diligence investigations. It includes a dedicated reasoning phase in which the model thinks through gathered material before generating its final output, which helps produce more nuanced and accurate responses. The model does not use customer queries or outputs for training purposes. It is well-suited for professionals, researchers, and developers working in domains like finance, technology, healthcare, and current events who need reliable, well-sourced reports.

What Sonar Deep Research supports

Multi-Step Web Search

Conducts dozens of iterative web searches per query, evaluating and refining results across each step to build a comprehensive picture of a topic.

Source Synthesis

Combines findings from hundreds of sources into a single coherent report, with citations included throughout the output.

Deep Reasoning

Applies a dedicated reasoning phase before generating a final response, allowing the model to evaluate gathered material and produce more accurate, nuanced outputs.

128K Context Window

Supports up to 128,000 tokens per session, enabling large volumes of text, citations, and research material to be processed together.

Cited Report Generation

Produces structured, long-form reports with inline citations linking back to the original web sources consulted during research.

Autonomous Research Planning

Independently determines a research strategy for a given query, deciding which sources to consult and how to iterate without requiring user guidance at each step.

Ready to build with Sonar Deep Research?

Get Started Free

Benchmark scores

Scores represent accuracy — the percentage of questions answered correctly on each test.

BenchmarkWhat it testsScore
MMLU-ProExpert knowledge across 14 academic disciplines68.9%
GPQA DiamondPhD-level science questions (biology, physics, chemistry)47.1%
MATH-500Undergraduate and competition-level math problems81.7%
AIME 2024American math olympiad problems48.7%
LiveCodeBenchReal-world coding tasks from recent competitions29.5%
HLEQuestions that challenge frontier models across many domains7.3%
SciCodeScientific research coding and numerical methods22.9%

Common questions about Sonar Deep Research

What is the context window for Sonar Deep Research?

Sonar Deep Research supports a context window of 128,000 tokens, allowing large amounts of text, citations, and research content to be handled within a single session.

How is Sonar Deep Research different from a standard search or retrieval model?

Rather than returning a single retrieved answer, Sonar Deep Research autonomously plans a research strategy, performs dozens of iterative web searches, evaluates sources, and refines its approach before producing a detailed, citation-backed report.

Does Perplexity use my queries or outputs to train Sonar Deep Research?

No. According to the model's documentation, customer queries and outputs are not used to train the model.

What is the knowledge cutoff or training date for Sonar Deep Research?

The model's training data has a cutoff of February 2025. However, because it performs live web searches at inference time, it can access and cite information published after that date.

Where can I find pricing information for Sonar Deep Research?

Pricing details are available on Perplexity's official pricing page at https://docs.perplexity.ai/docs/getting-started/pricing.

What types of tasks is Sonar Deep Research best suited for?

It is designed for tasks that require thoroughness over speed, including academic research, market analysis, competitive intelligence, and due diligence investigations across domains like finance, technology, and healthcare.

Parameters & options

Max Temperature1.9
Max Response Size8,000 tokens
Return CitationsSelect

Determines whether or not a request to an online model should return citations.

Default: false
NoYes
Return ImagesSelect

Determines whether or not a request to an online model should return images.

Default: false
NoYes

Start building with Sonar Deep Research

No API keys required. Create AI-powered workflows with Sonar Deep Research in minutes — free.