Pick the cheapest Perplexity Sonar model that still meets an agent task's depth-of-research and reasoning needs.

Documented steps

1. Read the model family overview at https://docs.perplexity.ai/docs/sonar/models, which lists exactly four current models: sonar, sonar-pro, sonar-reasoning-pro, sonar-deep-research.
2. Use plain sonar for quick, single-hop factual lookups with light search grounding — the cheapest tier at $1 per 1M input/output tokens (https://docs.perplexity.ai/docs/getting-started/pricing).
3. Use sonar-pro when a query needs multiple sources synthesized into one grounded answer; priced at $3 input / $15 output per 1M tokens.
4. Use sonar-reasoning-pro when the task needs chain-of-thought-style reasoning over search results (comparisons, multi-step logic); priced at $2 input / $8 output per 1M tokens.
5. Use sonar-deep-research for exhaustive, multi-source research — docs describe it searching hundreds of sources with a 128K context window, billing separately for reasoning tokens (~$3/1M) and search queries (~$5/1K) on top of input/output tokens (https://docs.perplexity.ai/docs/sonar/models/sonar-deep-research).
6. Because sonar-deep-research responses can take a long time, pair it with the async job pattern (POST /v1/async/sonar) instead of a single blocking synchronous call.