Tool profile
Perplexity
Perplexity is an AI search engine that answers questions in plain English while citing its sources in real time. Unlike a traditional search engine, it reads the web for you and synthesizes a direct answer rather than returning a list of links. It can handle follow-up questions, run deep research reports, and search within specific domains.
Models and pricing
Casual queries, basic web search, everyday factual questions
Uses a mix of smaller models and limited access to more powerful ones. You get a handful of Pro searches per day before being throttled to the standard tier. Good enough for most simple lookups.
Heavy research use, longer reports, access to the best underlying models
Unlocks unlimited Pro searches, lets you choose which underlying AI model powers your answer (GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, or Perplexity's own models), and adds file upload and image generation. Pick this if you use Perplexity daily or need reliable access to frontier models for research.
Developers building lightweight search-augmented apps
A smaller, faster model with live web access baked in. Cheap enough for high-volume apps where you need real-time web grounding without paying for the heavy model.
Developers who need stronger reasoning plus live web access
More capable than sonar-small, still cheaper than calling frontier models directly through OpenAI or Anthropic. Use this when answer quality matters and you can absorb a higher per-query cost.
Teams and organizations needing SSO, admin controls, and data privacy guarantees
Adds team management, the ability to connect internal data sources, and stronger data-handling commitments. Contact Perplexity directly for a quote. Not worth it for individuals or small teams.
Which one should you use?
Most users should start with the free tier to see if the workflow fits, then upgrade to Pro at $20/month if they find themselves hitting the daily Pro search limit or wanting to choose between Claude, GPT-4o, and Gemini for their answers. Developers wanting live-web-grounded responses in their own apps should look at the sonar API models.