# Sonar

Sonar retrieves live web results and synthesizes them into cited answers in a single API call. You don't need external search tool chains.

- **Model ID:** `perplexity/sonar`
- **Type:** chat
- **Providers:** perplexity
- **Context window:** 127,000
- **Maximum output tokens:** 8,000
- **Pricing:** $0.25/1M input tokens, $2.50/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/sonar

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'perplexity/sonar',
  prompt: 'Why is the sky blue?'
})
```

## About

Sonar is the base tier of Perplexity's Sonar API family. It's built for developers who need web-grounded answers without building a retrieval pipeline. Unlike conventional language models that generate from training data alone, Sonar models run a live web search on every inference call and synthesize results into a cited response.

This built-in search is Perplexity's core differentiator. You don't need to configure a separate search API, build a RAG pipeline, or manage document indexing. The model handles query formulation, web retrieval, source evaluation, and answer synthesis in one request. Responses include inline citations pointing to sources, giving your application verifiable provenance for each claim.

Sonar targets high-volume use cases where cited answers matter but deep reasoning isn't required. It has the lowest per-token cost in the Sonar family. That makes it practical for customer-facing search, chatbots that need factual accuracy, and automated research assistants processing many queries. The context window of 127K tokens supports multi-turn conversations that accumulate search context across exchanges.

## What to consider

Sonar runs a web search on every request. Response latency includes network round-trips to retrieve sources. Factor this into timeout budgets for latency-sensitive interfaces. Web search calls bill separately ($2.5 per thousand when listed for this model).

## When to use

### Best For

- **Customer-facing search** where answers must cite current web sources
- **High-volume factual Q&A** at the lowest cost in the Sonar family, suitable for chatbots and support assistants
- **Automated research pipelines** that gather and synthesize web information across many queries without separate search infrastructure
- **Content verification workflows** where citation-backed responses provide auditable source trails
- **Lightweight RAG replacement** without the complexity of building and maintaining a custom retrieval stack

### Consider Alternatives When

- The task requires deep multi-step reasoning over search results. Sonar Reasoning Pro applies chain-of-thought before answering
- You need higher search quality and more thorough source analysis. Sonar Pro searches more sources and produces more detailed synthesis
- The workload is purely generative (creative writing, code generation) with no need for web grounding
- You're in an offline or air-gapped environment where web search calls can't run

## Best for

- **Customer-facing search:** Answers must cite current web sources
- **High-volume factual Q&A:** The lowest cost in the Sonar family, suitable for chatbots and support assistants
- **Automated research pipelines:** Gather and synthesize web information across many queries without separate search infrastructure
- **Content verification workflows:** Citation-backed responses provide auditable source trails
- **Lightweight RAG replacement:** No need to build and maintain a custom retrieval stack

## Consider alternatives

- **Deep multi-step reasoning:** Sonar Reasoning Pro applies chain-of-thought before answering
- **Thorough source analysis:** Sonar Pro searches more sources and produces more detailed synthesis
- **Purely generative work:** Creative writing or code generation has no need for web grounding
- **Offline environments:** Air-gapped deployments can't make web search calls

## Frequently asked questions

### How does Sonar's built-in web search work?

Every API call triggers a live web search. The model formulates search queries from your prompt, retrieves and evaluates web sources, then synthesizes the information into a response with inline citations. You don't need an external search API or RAG pipeline.

### What is the difference between Sonar and Sonar Pro?

Sonar Pro searches more sources, produces longer and more detailed answers, and handles more complex multi-source queries. Sonar is optimized for speed and cost on straightforward factual queries.

### Does Sonar include citations in its responses?

Yes. Responses include inline citations referencing the web sources used for each claim. This gives your application verifiable provenance.

### What is the context window for Sonar?

127K tokens. This supports multi-turn conversations where search context accumulates across exchanges.

### How much does Sonar cost?

Current pricing is shown on this page. AI Gateway routes across providers, and rates may vary by provider.

### How do I authenticate with Sonar through AI Gateway?

Use your AI Gateway API key with the model identifier ``perplexity/sonar``. AI Gateway handles provider routing and authentication. You don't need a separate Perplexity API key when using gateway-managed access.

### Can I use Sonar without the web search feature?

No. Web search is integral to the Sonar architecture and runs on every request. If you need a model without search, use a general-purpose language model from another provider.

## Links

- [Model page](https://vercel.com/ai-gateway/models/sonar)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://sonar.perplexity.ai)
- [Provider pricing](https://docs.perplexity.ai/docs/pricing)