ModelsSonar Pro Guide: What the Higher Price Actually Buys
Sonar Pro lists at three times Sonar's input price and fifteen times its output price, yet real bills differ far less. Where the money goes and when Pro is worth it.
On the rate card, Sonar Pro looks like a different class of product from Sonar. Perplexity lists it at $3 per million input tokens and $15 per million output, against $1 and $1. Fifteen times the output price reads like a decision you should agonise over.
In practice the bill rarely moves that much, and the reason says a lot about how to use a search model well. This guide covers what Sonar Pro is, where its cost really comes from, the features that are specific to it, and what Perplexity's 2026 API changes mean for anyone choosing it today.
What Sonar Pro is for
Sonar Pro is Perplexity's step up from Sonar for questions that need more than one fact: comparing sources, following up on a first result, reconciling claims that disagree. The description on its OpenRouter listing pitches it at in-depth, multi-step queries. It has been listed on OpenRouter since March 2025, with text and image input.
The OpenRouter listing gives it a 200,000-token context window and a maximum output of 8,000 tokens. That shape is telling. It is built to read a lot and write a little: take in many sources, answer concisely. If you want a long report, Sonar Pro's output cap will stop you, and Sonar Deep Research is the model designed for that.
Where the money goes
Perplexity charges Sonar Pro a request fee on top of tokens: $6, $10 or $14 per thousand calls at low, medium and high search context. Sonar's equivalent fees are $5, $8 and $12. The fee gap is small; the token gap is large. Which one dominates depends on your traffic.
A worked example, using published rates. Ten thousand questions, each sending about 2,000 input tokens and getting about 400 back, at medium search context:
- Sonar Pro: 20M input tokens at $3 is $60, 4M output at $15 is $60, request fees $100. About $220.
- Sonar: 24M tokens at $1 is $24, request fees $80. About $104.
Roughly twice the cost, not fifteen times. The multiplier grows as answers get longer, because output is where Pro's premium sits. So the cheapest way to run Sonar Pro is to ask for short answers: instruct it to be concise and set max_tokens. Input vs output token pricing covers why output-heavy pricing punishes verbose prompts more than long inputs.
Pro Search and the search_type switch
In November 2025 Perplexity made Pro Search generally available for Sonar Pro. OpenRouter's listing calls it Perplexity's most advanced agentic search system: rather than one retrieval pass, the model can run several searches before answering. Perplexity also added search_type: "auto", which lets its system choose the search path per query based on complexity.
On OpenRouter, Pro Search is exposed as a separate model id with its own per-request search price. The detail is in the Sonar Pro Search guide. The short version: it costs more per call, and it is worth it only on questions where a single search demonstrably misses.
Features specific to Sonar Pro
Perplexity's changelog lists one capability as exclusive to sonar-pro: a Media Classifier, added in December 2025, which automatically adds visual content to responses for qualifying queries. If your interface can render images alongside answers, that is a reason to choose Pro that has nothing to do with accuracy. If it cannot, ignore it.
One thing Sonar Pro does not have is a separate reasoning mode. Perplexity sells that as Sonar Reasoning Pro, which is cheaper per output token and listed with a reasoning trace. For analytical questions over search results, compare the two directly before assuming Pro is the stronger choice.
One preset for two prices
Perplexity says Sonar Chat Completions support ended on 27 September 2026. Synchronous and streaming requests keep working and are being reformulated as requests to its newer Agent API, rolled out model by model; asynchronous requests are no longer supported.
The interesting detail is the mapping. Perplexity's migration guide maps both Sonar and Sonar Pro to the same Agent API fast preset, and describes that preset as much more accurate than Sonar Pro, on Perplexity's internal benchmarks. Read plainly, the vendor no longer treats Sonar Pro as a distinct capability tier. The docs do not say whether the same applies to third-party routes, so if you call Sonar Pro through OpenRouter, rerun a fixed question set periodically to catch behaviour changes.
Deciding with a paired test
- Take 30 real questions that need more than one source to answer.
- Run each through Sonar and Sonar Pro at the same search context size.
- Score correctness, and whether each citation actually supports its claim.
- Divide total cost by correct answers for each model.
If Sonar Pro wins on cost per correct answer, use it. If Sonar matches it on most questions, route by difficulty: Sonar first, Pro for the question types where Sonar fails. Model routing and fallbacks covers how to set that up without doubling latency.
Common questions
How much more expensive is Sonar Pro than Sonar?
On the rate card, three times on input and fifteen times on output. In real workloads the per-request fee narrows the gap; with short answers Sonar Pro often costs about twice as much as Sonar per question, rising as answers get longer.
What is the context window of Sonar Pro?
OpenRouter lists 200,000 tokens of context with a maximum output of 8,000 tokens. It is designed to read many sources and answer concisely, not to write long reports.
What is Pro Search in Sonar Pro?
A multi-step search mode, generally available since November 2025, where the model runs several searches before answering. Perplexity also offers search_type auto to route queries by complexity; OpenRouter exposes Pro Search as a separate model id.
