Text Generation Model

Sonar Deep Research

Sonar Deep Research is a text generation model from Perplexity that performs multi-step web research and returns cited answers.

Start Building with Sonar Deep Research View All Models

Publisher

Perplexity

TypeText

Context Window128,000 tokens

ReleasedFebruary 2025

Input$2.00/MTok

Output$8.00/MTok

Try Sonar Deep Research →

About Sonar Deep Research

Deep research with real-time web citations

Sonar Deep Research is a large language model developed by Perplexity AI, released in February 2025. It is designed to conduct extended, multi-step research tasks by querying the web iteratively and synthesizing information into comprehensive, cited responses. The model supports a 128,000-token context window and can return citations and images alongside its text output.

The model is best suited for tasks that require gathering and consolidating information from multiple online sources, such as literature reviews, market research, technical investigations, and fact-heavy writing. Unlike standard chat models, Sonar Deep Research is built around the retrieval and attribution of real-time web content, making source transparency a core part of its output. It is available through MindStudio with configurable options for returning citations and images.

Capabilities

What Sonar Deep Research supports

Web-Grounded Research

Performs iterative, multi-step web searches to gather information before generating a response. This allows the model to synthesize content from multiple live sources in a single query.

Citation Return

Optionally returns source citations alongside generated text, allowing users to verify the origin of each claim. Configurable via the Return Citations input.

Image Return

Can return relevant images alongside text responses when the Return Images option is enabled. Images are sourced from the web during the research process.

Long Context Window

Supports a 128,000-token context window, enabling the model to process and reason over large documents or extended conversation histories.

Text Generation

Generates structured, detailed text responses with a maximum response size of 8,000 tokens, suitable for long-form reports and summaries.

Ready to build with Sonar Deep Research?

Get Started Free

Performance

Benchmark scores

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

Benchmark	What it tests	Score
MMLU-Pro	Expert knowledge across 14 academic disciplines	68.9%
GPQA Diamond	PhD-level science questions (biology, physics, chemistry)	47.1%
MATH-500	Undergraduate and competition-level math problems	81.7%
AIME 2024	American math olympiad problems	48.7%
LiveCodeBench	Real-world coding tasks from recent competitions	29.5%
HLE	Questions that challenge frontier models across many domains	7.3%
SciCode	Scientific research coding and numerical methods	22.9%

FAQ

Common questions about Sonar Deep Research

What is the context window size for Sonar Deep Research?

Sonar Deep Research supports a context window of 128,000 tokens, allowing it to handle large inputs and extended conversations.

What is the maximum response size?

The model has a maximum response size of 8,000 tokens per output.

Does Sonar Deep Research support image inputs?

Based on the available metadata, image input support is not confirmed for this model. It can, however, return images from the web as part of its research output when that option is enabled.

Can Sonar Deep Research return citations?

Yes. The model includes a configurable Return Citations option that, when enabled, includes source references alongside the generated response.

What is the knowledge cutoff for Sonar Deep Research?

Sonar Deep Research retrieves information from the live web at query time rather than relying solely on a fixed training cutoff, so its responses can reflect current information available online.

Is pricing information available for Sonar Deep Research?

Pricing details are not published in the available metadata. You can check MindStudio or Perplexity's official channels for current pricing information.

Resources