Perplexity Pro subscribers can choose from multiple large language models for each search. The default model Sonar is built by Perplexity, but you can also select GPT-4o, Claude 3.5 Sonnet, or Grok-2. Each model has different strengths, speed, and pricing implications. This article explains the differences between these four models and helps you decide which one to use for your specific task.
Key Takeaways: Choosing the Right Model in Perplexity Pro
- Settings > Model Selector (top-left of search bar): Switch between Sonar, GPT-4o, Claude 3.5 Sonnet, and Grok-2 for each query.
- Sonar (default): Fastest response time and lowest token cost; best for everyday factual searches.
- GPT-4o: Strongest for creative writing, code generation, and complex reasoning tasks.
- Claude 3.5 Sonnet: Excels in long-form analysis, document summarization, and nuanced explanations.
- Grok-2: Best for real-time data, trending topics, and X (formerly Twitter) integration.
Why Perplexity Offers Multiple Models
Perplexity Pro gives you access to four distinct large language models because no single model is best for every task. Sonar is the proprietary model optimized for speed and cost efficiency. GPT-4o from OpenAI provides broad general intelligence with strong creative and coding abilities. Claude 3.5 Sonnet from Anthropic focuses on safety, nuance, and handling long contexts. Grok-2 from xAI is tuned for real-time information and social media data.
Each model has different token limits, response speeds, and knowledge cutoff dates. Sonar uses a lightweight architecture that returns answers in 1-3 seconds. GPT-4o and Claude take 3-8 seconds but offer deeper reasoning. Grok-2 can access live data from X and is updated more frequently than the others.
Your choice also affects the number of Pro searches you consume. Sonar uses one search credit per query. GPT-4o and Claude use two credits. Grok-2 uses one credit. If you have a limited number of Pro searches per day, Sonar or Grok-2 may be more economical.
Knowledge Cutoff Dates
Sonar has a knowledge cutoff of April 2024. GPT-4o has a cutoff of October 2023. Claude 3.5 Sonnet has a cutoff of April 2024. Grok-2 has a cutoff of August 2024 and can also pull live data from X and web searches. For very recent events, Grok-2 is the best choice.
Context Window Sizes
Sonar supports a 128K token context window. GPT-4o supports 128K tokens. Claude 3.5 Sonnet supports 200K tokens. Grok-2 supports 128K tokens. Claude can process the longest documents in a single query, making it ideal for analyzing large PDFs or long articles.
Steps to Switch Models in Perplexity Pro
- Open Perplexity and sign in to Pro
Go to perplexity.ai and log in with your Pro account. The model selector is only available to Pro subscribers. - Locate the model selector dropdown
In the search bar area, look for a small dropdown button at the top-left side of the input field. It shows the current model name, such as Sonar. - Click the dropdown to see all models
A list appears with Sonar, GPT-4o, Claude 3.5 Sonnet, and Grok-2. Each entry shows its search credit cost per query. - Select the model for your query
Click any model to activate it. The search bar updates to show the selected model. Type your question and press Enter. - Verify the model in the response
After the answer appears, the model name is displayed at the top of the response card. You can switch models mid-session without losing your conversation history.
Common Mistakes When Choosing a Model
I Selected GPT-4o but Got a Short Answer
GPT-4o may give concise answers if the query is simple. For detailed output, add explicit instructions such as Explain this in detail or Write a 500-word analysis. The model follows user prompts closely.
Sonar Returns Outdated Information for Breaking News
Sonar has an April 2024 cutoff. For current events, use Grok-2 which can search live web and X data. You can also enable web search in Sonar by clicking the globe icon next to the search bar, but the model itself still relies on its training data for synthesis.
Claude Refuses to Generate Code or Creative Content
Claude has stricter safety filters than other models. If Claude blocks a request, switch to GPT-4o or Grok-2 for coding tasks. This is by design to reduce harmful outputs.
Grok-2 Fails to Find Information Not on X
Grok-2 prioritizes X data. For general web information, use Sonar or GPT-4o. Grok-2 also has limited support for non-English languages compared to GPT-4o.
Perplexity Pro Model Comparison
| Item | Sonar | GPT-4o | Claude 3.5 Sonnet | Grok-2 |
|---|---|---|---|---|
| Provider | Perplexity | OpenAI | Anthropic | xAI |
| Knowledge cutoff | April 2024 | October 2023 | April 2024 | August 2024 + live data |
| Context window | 128K tokens | 128K tokens | 200K tokens | 128K tokens |
| Search credits per query | 1 | 2 | 2 | 1 |
| Average response time | 1-3 seconds | 3-6 seconds | 4-8 seconds | 2-5 seconds |
| Best for | Fast factual answers | Creative writing, code, reasoning | Long document analysis, nuanced topics | Real-time news, X data, trends |
| Multimodal support | No | Yes (image input) | Yes (image input) | No |
| Language support | 95+ languages | 95+ languages | 95+ languages | English dominant |
You can now switch between Sonar, GPT-4o, Claude 3.5 Sonnet, and Grok-2 based on your task requirements. For daily quick searches, stay with Sonar to save credits. For complex coding or creative projects, use GPT-4o. For analyzing long documents, choose Claude. For the latest news or X trends, select Grok-2. A useful advanced technique is to combine models: use Sonar for the initial answer, then switch to GPT-4o to expand or refine the response without losing context.