Sonar Deep Research

Name: Sonar Deep Research
Author: Perplexity

PerplexityProprietaryPending Human Review

Sonar Deep Research was Perplexity’s separately named API model tier for extended research reports. It repeatedly searched and analyzed sources to produce a cited synthesis, with adjustable reasoning effort and synchronous or asynchronous research workflows. The Deep Research application was announced in February 2025, but this page dates the verifiable named API identity to March rather than assuming the app used the same model from launch. Its underlying model architecture and parameter count were not disclosed. Sonar Chat Completions support ended September 27, 2026; the provider maps this tier to Agent API research presets, and legacy asynchronous requests are no longer supported.

2025-03

Undisclosed language-model architecture with iterative search and reasoning

Proprietary

Compare with other models

Specifications

Architecture: Undisclosed language-model architecture with iterative search and reasoning
License: Proprietary
Context Window: 128,000 tokens
Type: text
Modalities: text

Benchmark Scores

Advanced Specifications

Model Family: Sonar
API Access: Not Available
Chat Interface: Not Available
Variants: sonar-deep-researchreasoning_effort settings

Capabilities & Limitations

Capabilities: iterative researchmulti-step reasoningcited reportsadjustable effort
Known Limitations: Retired API tier is mapped to Agent API presetsLegacy asynchronous Sonar requests are unsupportedFebruary app launch does not verify the named API model was available thenReports require checking source quality and claims
Notable Use Cases: research reportsliterature discoverymarket and technical synthesis
Tool Use Support: Yes

Resources

Related Models

Sonar Reasoning Pro

Perplexity

Sonar Reasoning Pro was the higher-capability reasoning tier in Perplexity’s Sonar API, combining chain-of-thought reasoning with deeper web retrieval and more citations for multi-step questions. The distributor describes it as DeepSeek R1-powered, while the exact internal post-training recipe remains undisclosed. Public March 2025 provider documentation and serving records establish availability that month; a precise earliest launch day was not recovered. Search-effort settings belong to this model rather than separate pages. The original service ended support September 27, 2026, and compatibility requests are gradually reformulated as Agent API requests using presets, which does not establish continued serving of this historical model.

R1 1776

Perplexity

R1 1776 is Perplexity’s significant post-training of DeepSeek R1, targeting refusals and evasive responses around politically sensitive Chinese topics while retaining reasoning capabilities. It is an offline language model and does not use the Sonar search subsystem. The original publisher announcement and model repository have since become inaccessible, but the release is corroborated by the publisher’s February 18 communication, distributor records, and preserved model-card copies and weights. The MIT-licensed 671B checkpoint supports local deployment with substantial hardware resources. Perplexity removed the hosted model on August 1, 2025. Claims of being unbiased are publisher positioning rather than an established guarantee.

Sonar Pro

Perplexity

Sonar Pro was Perplexity’s separately positioned premium search model, providing deeper retrieval and more sources for complex questions than Sonar. Its hosted product combined language generation with web grounding, search filters, citations, and later image-input support. High, Medium, and Low search settings and the later Pro Search mode are options within this identity. Internal architecture and parameter counts were not disclosed sufficiently to identify a particular backbone. Sonar Chat Completions support ended September 27, 2026; the provider now maps this historical tier to Agent API presets, so continued compatibility requests do not verify continued inference on Sonar Pro itself.