Text Generation ModelDeprecated

Sonar Reasoning

Sonar Reasoning is a text generation model from Perplexity combining chain-of-thought reasoning with live internet search and citations.

Start Building with Sonar Reasoning View All Models

Publisher

Perplexity

TypeText

Context Window127,000 tokens

Replaced bySonar Reasoning Pro

Try Sonar Reasoning →

About Sonar Reasoning

Real-time search with chain-of-thought reasoning

Sonar Reasoning is a text generation model developed by Perplexity that combines chain-of-thought reasoning with real-time internet search. It is built on reasoning capabilities trained with DeepSeek R1 and is hosted in US-based datacenters. The model supports a context window of up to 127,000 tokens and can return responses up to 32,768 tokens in length.

What distinguishes Sonar Reasoning from standard chat models is its ability to retrieve current web information and attach citations to its responses, making it suited for tasks that require up-to-date facts alongside structured reasoning. The model is described as uncensored and is configurable with options for returning citations, returning images, and adjusting search context size. It is noted as deprecated in the MindStudio catalog, meaning it has been succeeded by newer versions in Perplexity's model lineup.

Capabilities

What Sonar Reasoning supports

Chain-of-Thought Reasoning

Applies step-by-step reasoning trained with DeepSeek R1 to work through complex queries before producing a final answer.

Real-Time Web Search

Retrieves live information from the internet at inference time, allowing responses to reflect current events and data.

Citation Return

Optionally attaches source citations to responses, configurable via the Return Citations input parameter.

Image Results

Can return image results alongside text responses when the Return Images parameter is enabled.

Search Context Control

Allows adjustment of how much web search context is incorporated into a response via the Search Context Size parameter.

Large Context Window

Supports up to 127,000 tokens of context, enabling long documents or extended conversation histories in a single request.

Ready to build with Sonar Reasoning?

Get Started Free

FAQ

Common questions about Sonar Reasoning

What is the context window size for Sonar Reasoning?

Sonar Reasoning supports a context window of 127,000 tokens, with a maximum response size of 32,768 tokens.

Is Sonar Reasoning still available to use?

Sonar Reasoning is marked as deprecated in the MindStudio catalog, which means it has been succeeded by newer models in Perplexity's lineup. Availability may be limited or discontinued.

What makes Sonar Reasoning different from a standard chat model?

Sonar Reasoning combines chain-of-thought reasoning trained with DeepSeek R1 with real-time internet search, allowing it to reason through problems while also retrieving and citing current web information.

Does Sonar Reasoning support image inputs?

Based on the available metadata, image input support is not confirmed for Sonar Reasoning. However, it does have a Return Images parameter that can include image results in its output.

Where is Sonar Reasoning hosted, and is there a pricing model?

Sonar Reasoning is hosted in US-based datacenters operated by Perplexity. No specific pricing information is available in the current metadata.

Resources