​
Perplexity AI Models

The following Perplexity AI models are available. Each model offers distinct capabilities tailored for various applications and needs.
ModelsGenerate TextWeb SearchReasoningResearch DepthContext WindowMax Output TokensAvailability
sonar-pro✅✅❌High200k tokens8k tokensPro
sonar✅✅❌Medium128k tokens-Free
Please note that some features may not be available within Giselle even if they are offered in the official API.

​
sonar-pro

An advanced search model offering detailed responses with enhanced citation accuracy. Optimized for multi-step queries and tasks that require deep content understanding and comprehensive sourcing. Ideal for academic research, market analysis, and complex informational tasks.

​
sonar

A lightweight, cost-effective search model optimized for quick, accurate answers utilizing real-time web search with citations. Suitable for everyday queries, quick fact-checking, summarizations, and general-purpose use.

​
Model Selection Guide

Guidelines for selecting the optimal Perplexity AI model:
  • For advanced multi-step research queries with citations: sonar-pro
  • For everyday quick searches and cost-effective queries: sonar

​
Practices for Giselle

​
Multi-Node Research

In Giselle, you can set up multiple sonar-pro nodes to approach different perspectives, creating detailed, multifaceted research outputs similar to original deep-research reports.

​
Search Domain Filtering

Both sonar and sonar-pro nodes in Giselle now support the search_domain_filter option, allowing you to control which websites are included in or excluded from the search results used by the models. This feature is particularly useful for:
  • Restricting search results to trusted sources.
  • Filtering out specific domains (e.g., forums, social media).
  • Focusing research on particular websites.
How to Use:
  • Allowlist (Include): Provide a list of simple domain names (e.g., "wikipedia.org", "nasa.gov"). The search will only use results from these domains.
  • Denylist (Exclude): Provide a list of simple domain names (e.g., "pinterest.com", "reddit.com"). The search will exclude results from these domains.
  • Combine: You can mix included and excluded domains in the same request.
Best Practices:
  • Simple Domain Names: Use domains in their simplest form (e.g., example.com) without protocol prefixes (http://, https://) or www..
  • Main Domains: Using the main domain (e.g., nytimes.com) will filter all its subdomains as well.
  • Limit Filter Size: You can add a maximum of 10 domains. Using fewer, more targeted domains often yields better results.
  • Relevance: Use domains most relevant to your query for the best results.
Performance Considerations:
  • Adding domain filters might slightly increase response time.
  • Overly restrictive filters could lead to fewer search results, potentially affecting response quality.
For detailed specifications or additional assistance, please refer to Perplexity AI Documentation.