Perplexity (Sonar) API Pricing
Sonar models are search-grounded, which makes their token rates genuinely misleading on their own. There is a second meter running: a per-request search fee that can easily exceed the token cost.
| Model | Input / 1M | Output / 1M | Cached input / 1M | Batch | Notes |
|---|---|---|---|---|---|
| Sonar Small Online | $0.20 | $0.20 | N/A | N/A | Cheapest Sonar tier. Per-request search fees apply on top. |
| Sonar | $1.00 | $1.00 | N/A | N/A | Flat $1 in / $1 out. Per-request search fees apply on top. |
| Sonar Huge Online | $5.00 | $5.00 | N/A | N/A | Flat input/output rate. Per-request search fees apply on top. |
| Sonar Pro | $3.00 | $15.00 | N/A | N/A | Search-grounded model. Token rates are only part of the bill: Sonar models also charge a per-request search fee of roughly $5-$14 per 1,000 requests depending on search depth. |
Perplexity's cheapest tracked model is Sonar Small Online at $0.20 in / $0.20 out, and its most expensive is Sonar Pro at $3.00 in / $15.00 out, a 30x spread across the lineup. Sonar Pro ranks #54 of 60 models tracked across 17 providers.
The search fee is the part people miss
Every Sonar model charges roughly $5 to $14 per 1,000 requests for the search itself, depending on the model and how deep the search goes, on top of the per-token rates in the table above. At $10 per 1,000 requests, that is one cent per call before a single token is billed.
For short queries, that fee dominates completely. A request with 1,000 input and 300 output tokens on Sonar costs about $0.0013 in tokens and around $0.01 in search: roughly eight times more for the search than for the inference. Any cost model for Sonar that only counts tokens will be wrong by close to an order of magnitude, and the error gets worse the shorter your requests are.
Comparing Sonar to a plain LLM is not like-for-like
Sonar is doing retrieval, not just generation. The honest comparison is against a model plus your own search infrastructure, plus the input tokens the retrieved documents consume, not against a bare model call. Built that way, a search-grounded API frequently comes out ahead, because the retrieved-context tokens you would otherwise pay for on a general model are substantial.
The RAG pipeline cost estimator is the right tool for that comparison: it prices the retrieval side rather than assuming it is free.
The tier ladder
Sonar and Sonar Huge Online charge the same rate for input and output, which is unusual and quietly favourable for generation-heavy work: on most providers output runs four to six times input. Sonar Pro reverts to the conventional shape at a five-to-one output premium. Sonar Small Online is the volume tier.
Every Perplexity model on this site
| Model | Blended / 1M | Rank of 60 | Full pricing |
|---|---|---|---|
| Sonar Small Online | $0.200 | #11 | Sonar Small Online pricing page |
| Sonar | $1.00 | #23 | Sonar pricing page |
| Sonar Huge Online | $5.00 | #50 | Sonar Huge Online pricing page |
| Sonar Pro | $6.00 | #54 | Sonar Pro pricing page |
Perplexity pricing FAQ
How much does the Perplexity Sonar API cost?
Token rates run from $0.20 to $5 per 1M input depending on tier, but every request also carries a search fee of roughly $5 to $14 per 1,000 requests. For short queries that fee is the larger of the two costs by a wide margin.
Why do Sonar models charge the same for input and output?
On the Sonar and Sonar Huge Online tiers, input and output are billed at a flat identical rate. That is unusual: most providers price output four to six times higher than input. It makes those tiers comparatively attractive for generation-heavy work and comparatively poor for prompt-heavy work.
Is Sonar cheaper than building my own RAG pipeline?
Often, once you count what your own pipeline actually costs: embedding, vector storage, retrieval, and the retrieved-context input tokens you feed to a general model. Compare like for like with the RAG pipeline cost estimator rather than against a bare model call.