Perplexity gives you access to the latest AI models from OpenAI and Anthropic, as well as other companies and includes both proprietary and open sourced models.
Here’s a quick overview of each model currently included—the list evolves as models and search modes update.
Search modes and models
Pro Search
Pro search uses your default or baseline model and performs an in depth internet search, granting you 3x the amount of sources. You can pick an advanced model when performing a Pro Search of your choice. The latest advanced models we support are:
"Best" mode
“Best” mode is optimized for speed, browsing, and accuracy, automatically selecting the ideal model for your query to ensure top-tier results every time.
Pro Search Models
Sonar
Powered by Llama 3.1 70B and refined for Perplexity’s real-time search engine. Sonar specializes in fast web search and summarization, with the option to enable advanced reasoning when your query requires deeper analysis.
GPT-5.2
OpenAI’s latest model excels in reasoning, coding, and creativity with reduced hallucination rates. The reasoning option lets you toggle advanced logical processing, ideal for complex questions and tasks.
Claude Sonnet 4.6
Anthropic’s Sonnet 4.6 is renowned for its efficient coding and technical reasoning. Enable the reasoning toggle to unlock more thoughtful, analytical responses for technical tasks and challenging queries.
Gemini 3.1 Pro
Google’s newest AI offers state-of-the-art multimodal understanding and code generation—reasoning is always enabled, ensuring thorough, thoughtful analysis for every search.
Claude Sonnet 4.6 Thinking
The Sonnet 4.6 thinking model delivers notably stronger coding skills, optimizing efficiency for software development and technical problem-solving tasks.
Claude 4.6 Opus (only available to Perplexity Max and Enterprise subscribers)
Anthropic’s flagship model, engineered for demanding reasoning and agentic tasks. The reasoning toggle unlocks full depth for nuanced, complex problem solving (Perplexity Max and Enterprise Max users only).
Nemotron 3 Super 120B with Thinking (by NVIDIA)
NVIDIA's large-scale 120B parameter model with built-in reasoning capabilities, optimized for advanced problem solving and deep analytical tasks.
How do I choose the right model?
The best way to figure out which model works for you is to test them out. Try comparing your queries across different models and see which one suits your needs best. It all comes down to what you’re looking to achieve.
What’s a token?
A token is a small chunk of text that AI models break down to process. Think of tokens as building blocks that help AI understand and generate language. You can use a tokenizer to see how many tokens a given piece of text contains.
Are there any guidelines for using third-party models like Claude or GPT?
No, you can use these models just as you would in their original platforms.
How do I know which model to pick for my task?
While all up to date LLMs for the most part achieve similar performance on standard benchmarks, Perplexity's choice of models have distinct operational strengths optimized for web search use case.