Perplexity: Sonar Pro

Perplexity: Sonar Pro is a text generation model from Perplexity, released by the provider on 7 March 2025. It sits in the text category and is aimed at developers who need a hosted model that can answer queries with the option of web search behind it. Perplexity builds the model and serves it through its own API, and api14 exposes it through an OpenAI-compatible interface so you can call it without changing your existing client code.

The model accepts text and image inputs and returns text only. It has a context window of 200,000 tokens, which is large enough to hold long documents, transcripts or multi-turn conversations alongside your instructions. The maximum output is 8,000 tokens, so responses are bounded even when the input is very long. Supported request parameters are frequency_penalty, max_tokens, presence_penalty, temperature, top_k, top_p and web_search_options. The web_search_options parameter is the notable one here, since it lets you control the search behaviour that Sonar Pro is built around.

Those capabilities point to a few practical uses. Research and question answering over current information are the obvious fit, because the model can combine a large context with web search rather than relying only on what it learned during training. You can feed it a long report, a set of documents or an image alongside a question and ask for a written answer. Summarisation, comparison and multi-step investigation tasks also work well, since the 200,000 token window lets you keep source material in the same request instead of chunking it. Image input means screenshots, diagrams or scanned pages can be part of the prompt.

To call it through api14, use the model id "perplexity/sonar-pro" with the OpenAI-compatible chat completions endpoint. Your request can set frequency_penalty, max_tokens, presence_penalty, temperature, top_k, top_p and web_search_options, and you can send text and image content in the messages array. Because the interface matches OpenAI's, existing SDKs and libraries generally work with only the base URL and model id changed.

One limit worth noting is the 8,000 token output cap, which may be tight if you need very long generated documents in a single response. If your task does not need web search, or if you need output modalities other than text, another model may fit better. For search-backed answers over long inputs, Sonar Pro is the appropriate choice.