// how to
How to Choose an AI Model in Perplexity (Best, Thinking, Council)
Leave the model on Best for everyday questions; Perplexity picks a suitable model for each one. On a paid plan, select the model menu in the bottom right of the question box, choose Custom and pick a model. Turn on Thinking for complex, multi-step questions. Free accounts can’t choose a model manually.
The short version
Most of the time, Best is the right choice. Perplexity picks a model for each question. Choosing a model yourself is for when you’ve noticed one handles a type of question better, or when you want a second opinion on an answer.
Step by step
- Start with Best. If no model is selected, the menu in the question box says Best.
- To pick a model (paid plans), select the model menu in the bottom right of the question box, choose Custom, then pick the model, along with reasoning and speed options.
- Turn on Thinking for complex or multi-step questions. It gives the model more time to reason, so answers take longer. Some models use it automatically.
- Try another model on the same answer. Under an answer, the Rewrite button generates a new answer with a model you choose.
- Check which model answered. A button under each answer shows the model, search mode and source mode used.
- Use Deep Research when you want a report; it chooses its own models.
Which option for which job
Perplexity’s own advice:
- Best: everyday questions.
- Thinking: complex reasoning, planning, coding or multi-step analysis.
- Deep Research: a full report rather than one answer.
- Model Council: when comparing several models’ views matters, such as investment research, a big purchase or cross-checking a fact. Max and Enterprise Max only.
The Help Center adds that the most reliable way to find your preference is to run the same question across a few models and compare.
Prompts to copy
Run the same prompt on two or three models and compare:
[Your question]. Answer in under 150 words, with sources, and end with your confidence level
(high, medium, low) and why.
A good test of reasoning:
I have [situation with numbers]. Work out [what you need], show each step,
and tell me which assumption matters most.
Example
A bookkeeper asks about a tricky sales tax situation across two states. Best gives a clear answer. She selects Rewrite with a different model and Thinking on; the second answer flags an exception the first missed. She opens the state’s own tax page to settle it.
Mistakes to avoid
- Assuming a model name means quality. Test on your own questions.
- Leaving Thinking on for simple questions. It’s slower without being better.
- Treating two models agreeing as proof. They can both rely on the same weak source. Check the citations.
Good to know
The model list changes as providers release new models and Perplexity retires old ones, and some models are available only on Max. On a free account you use Best plus a limited number of Pro Searches.
Related: use Pro Search and use Perplexity for free.
Frequently asked questions
Which models does Perplexity offer?
It changes often. On October 27, 2026, Perplexity's Help Center listed Search models from OpenAI, Anthropic, Google, xAI, Moonshot AI, Z.ai and NVIDIA, with some reserved for Max. The model menu in your account is always the current list.
Is Claude or GPT in Perplexity the same as in their own apps?
No. Perplexity says it combines these models with its own search, citations, tools and limits, so answers can differ from the provider's own app.
What is Model Council?
A feature that runs one question across several models and combines their answers, showing where they agree and disagree. It's included on the Max and Enterprise Max plans, on the web and in the iOS app.
Free download
Get the AI Coding Starter Kit
A PDF you can use with any AI coding assistant: a project-rules template, 25 tested prompt patterns, a team adoption checklist and an assistant comparison sheet. The editable Markdown sources are included.
- Works with Copilot, Claude Code, Codex, Cursor and more
- Double opt-in, one-click unsubscribe
- No trackers, never shared