Sonar Deep Research: What It Does and Costs
perplexity/sonar-deep-researchA text-only model for multi-step research, source reading, synthesis, and reasoning.
128K tokens of context — about 96,000 words, on a scale that ends at 2M.
What it can do
- Reasoning
- Takes text
- Gives text
Released 7 Mar 2025 · checked 3 Oct 2026.
About it
Sonar Deep Research is Perplexity’s research-focused language model. It handles multi-step retrieval, reads and evaluates sources, and refines its approach while working through complex topics. It accepts text and returns text, supports reasoning, and has a 128,000-token context window with up to 115,200 output tokens.
Review
In a line
Sonar Deep Research is a Perplexity model built for complex research tasks. It searches, reads, evaluates, and combines information across several steps.
What it's good at
The model suits work that needs retrieval, source evaluation, synthesis, and reasoning rather than a single short answer.
It works with text input and produces text output. See reasoning models for related options.
Price in plain terms
Input costs $2 per million tokens, and output costs $8 per million tokens. Tokens are small pieces of text counted for billing.
At the stated rates, reading a page of about 500 words costs about 0.13 cents. One chat turn with about 1,000 input tokens and 500 output tokens costs about 0.6 cents. A thousand such turns would cost about $6.00.
Context and limits
It can take up to 128,000 tokens in one request, so a large amount of text can fit into a single task. One answer can contain up to 115,200 tokens.
The model is text-only. Its listed strengths cover retrieval, reading, synthesis, and reasoning; no image capability is provided in these details.
Ways to use it
Use it for text-based research that involves several searches, source reviews, and combined conclusions. Its model identifier is perplexity/sonar-deep-research.
You can compare it with other entries in Perplexity models or review lower-cost choices in cheapest language models.
Verdict
Choose Sonar Deep Research when a task needs extended reading and step-by-step reasoning across complex material. Its large context can hold substantial input, and its text format fits written research workflows.
Look elsewhere if you need images or want to minimize output spending. The output rate is four times the input rate, so long generated reports can cost more than the source text they process.
Verdict
Sonar Deep Research fits readers who need a model to work through complicated, text-heavy questions rather than produce a quick reply. Its 128,000-token context gives room for extensive material, while the reasoning and source-evaluation focus supports multi-stage research. Budget for generated text: output tokens cost $8 per million, compared with $2 for input. It is not a match for image-based work because both its inputs and outputs are text. Consider the wider Perplexity model range if you need a different balance of cost or capability.
- Handles multi-step retrieval, synthesis, and reasoning
- Reads and evaluates sources while refining its approach
- 128,000-token context window
- Supports up to 115,200 output tokens
- Text input and output keep the format straightforward
- Text-only; no image input or output
- Output costs $8 per million tokens
- Long generated answers can raise the bill
- Pick it ifComplex, multi-step text research; Tasks needing source reading and synthesis; Work that benefits from a large context window
- Skip it ifYou need image understanding or image generation; You mainly create long outputs on a tight budget; Your task does not need multi-step research
Pricing
Billing is split between text sent to the model and text it generates: $2 per million input tokens and $8 per million output tokens. A typical 1,000-token-in, 500-token-out chat turn is estimated at about 0.6 cents.
Questions
What is Sonar Deep Research?
It is a Perplexity model for multi-step retrieval, source reading, synthesis, and reasoning across complex topics.
How much does Sonar Deep Research cost?
Input costs $2 per million tokens. Output costs $8 per million tokens.
How large is its context window?
It accepts up to 128,000 tokens in one request.
How much can it write in one answer?
Its maximum output is 115,200 tokens.
Does it support images?
No image capability is listed. Its inputs and outputs are text.
What does a typical chat turn cost?
A turn with about 1,000 input tokens and 500 output tokens costs about 0.6 cents. One thousand such turns cost about $6.00.