Perplexity's developer platform sells web-grounded answers three ways: the Sonar models, the newer Agent API that now absorbs Sonar, and a raw Search API. Below is how Sonar moved onto the Agent API, and how search and tool fees stack on top of tokens.
A Perplexity API bill has two parts: per-token charges for the model, plus a fee for every search or tool call. Sonar Chat Completions support ended on September 27, 2026, and synchronous and streaming Sonar requests are being reformulated as Agent API requests, so check Perplexity's live Agent API model list for current Sonar token rates before you budget.
| Model | Tier | Input / 1M | Output / 1M | Cache hit / 1M | Context | Notes |
|---|---|---|---|---|---|---|
| Sonar Pro | Specialist | — | — | — | 200K | Not separately priced on the Agent API: Sonar Chat Completions support ended Sep 27, 2026, Sonar Pro isn't on the Agent API model list, and Perplexity maps it to the Agent API "fast" effort level (billed at that setup's model token rates plus tool fees) |
| Sonar | Specialist | $0.25 | $2.50 | $0.0625 | 128K | Cheap web-grounded answers; now served via the Agent API (perplexity/sonar) after Sonar Chat Completions support ended Sep 27, 2026; billed per token plus per tool call (web_search $0.0025 per invocation, $0.001 with Fast Search) |
“—” means Perplexity doesn't publish a metered per-token price for that model.
| Model | Support chatbot 10,000 conversations × (2,000 input + 400 output tokens) | Document Q&A 1,000 questions × (30,000 input + 800 output tokens) | Coding agent session 4M input tokens (90% cache hits) + 150,000 output tokens |
|---|---|---|---|
| Sonar | $15.00 | $9.50 | $0.70 |
Standard rates per 1M tokens from data on this site. The coding-agent row assumes 90% of input is read from the prompt cache (full input price where a model has no cache rate). Cache-write fees, batch discounts and long-context surcharges listed in the notes are not included. Updated November 2, 2026.
Perplexity ended support for Sonar Chat Completions on September 27, 2026. Synchronous and streaming requests keep working because they are reformulated as Agent API requests, rolled out gradually by model; asynchronous Sonar requests are no longer supported and have to move to the Agent API's background mode. The migration guide maps sonar and sonar-pro to the fast effort level, sonar-reasoning-pro to low and sonar-deep-research to high. Requests take an input field and return a typed output array with separate message and search_results items, instead of messages and choices.
On the Agent API you pay the model's token rate plus a per-invocation fee for each tool the model calls: web_search, fetch_url, image_search, people_search, finance_search, and a code sandbox billed per session. Fast Search invocations cost less than standard web_search. There's no equivalent of Sonar's search_context_size; instead you cap how much retrieved content comes back with max_tokens and max_tokens_per_page on the web_search tool. Agents that search several times per answer can spend more on tool calls than on tokens, so log invocations per request while testing.
The Agent API also routes to Claude, GPT, Gemini and Grok models at the per-token rates Perplexity lists for each, plus the same tool fees; only the OpenAI models offer a flex tier at 0.5× and priority at 2× the token price. Perplexity also hosts open-weight models, including Kimi K3 ($15.00 per 1M output) and GLM-5.3 ($4.40), at the same list rates as their makers. That lets one account and one bill cover search-grounded agents across several model families.
If you only need ranked web results for your own model, the Search API bills per successful request — up to five queries in one request count as a single billing unit, and failed, invalid or rate-limited requests are free — with a cheaper Fast Search mode. Agent API limits follow usage tiers set by lifetime credit purchases, from 1 query per second at Tier 0 to 33 at Tiers 4 and 5, with separate per-model token budgets per minute and per hour. A tier, once reached, is kept.
Sonar is billed per token plus a fee for each search or tool call. Since Sonar moved onto the Agent API in September 2026, the authoritative per-token rate is the one on Perplexity's Agent API models page, which lists sonar but not sonar-pro (the migration guide maps sonar-pro requests to the fast effort level). Check that page before you budget.
Yes. Support ended on September 27, 2026. Synchronous and streaming calls are being reformulated as Agent API requests; asynchronous calls must move to background mode.
Nothing with named sizes. On the Agent API you set explicit token budgets — max_tokens and max_tokens_per_page — on the web_search tool.
Agent API limits depend on a usage tier based on cumulative credit purchases, from 1 query per second at Tier 0 up to 33 at the top tiers, plus per-model token-per-minute and token-per-hour budgets. The Search API has a separate limit that doesn't depend on tier.
Yes. The Agent API serves Anthropic, OpenAI, Google and xAI models at per-token rates Perplexity lists for each, alongside its own Sonar model and hosted open-weight models.
How we choose: prices come from each vendor's official pricing page and are re-checked every week. We take no affiliate fees, so rankings aren't for sale.