Perplexity Sonar
6.3/10Perplexity's search-grounded language model, offered to developers through an API that returns current web answers with citations for apps and AI agents.
Data last updated November 2026 · Our methodology
Overview
Perplexity Sonar is Perplexity's grounded search model, which developers call over an API. Each request runs a live web search and returns an answer along with its sources, so teams can add cited, up-to-date answers to an app without building their own retrieval pipeline. Sonar originally came as a family of Chat Completions models: Sonar, Sonar Pro, Sonar Reasoning Pro and Sonar Deep Research. Perplexity ended support for Sonar Chat Completions on September 27, 2026. Sonar is now sold as the perplexity/sonar model inside Perplexity's Agent API, and the old model tiers map to effort levels: fast, low and high. Synchronous and streaming Sonar calls are being reformulated as Agent API requests automatically. Asynchronous calls are no longer supported and have to move to background mode. Pricing is pay-as-you-go. Sonar on the Agent API costs $0.25 per million input tokens and $2.50 per million output tokens. Tools are billed separately, for example web search at $2.50 per 1,000 calls and URL fetches at $0.0005 each. There is no ongoing free tier. The API follows an OpenAI-style format and works with tools such as MCP, sandboxes and custom functions. Gateways such as TrueFoundry and LLM Gateway also offer it. Perplexity says it applies zero data retention to API traffic and holds a SOC 2 Type II report. Sonar is best suited to developers building research features, support bots and agents that need fresh facts. Its main limitations are cost forecasting and churn: token charges plus per-tool fees are hard to predict, and the Sonar-branded endpoint has just been retired.
Our Rating
Who Perplexity Sonar is best for
Best software for DevelopersGood fit
Sonar is an API product built for developers who want to embed cited web answers in their own software.
Pay-as-you-go billing and low Sonar token rates let small teams add grounded search without running their own infrastructure.
It suits in-app research and support features, though costs at very high call volumes need watching.
Less suited
Zero data retention and SOC 2 Type II help, but the recent forced migration and limited control over retrieval are risks for large deployments.
Pros & Cons
Pros
- Returns answers grounded in live web results with source citations, so teams don't need to build a retrieval pipeline.
- Sonar's token rates on the Agent API ($0.25 input and $2.50 output per million tokens) are low for a search-grounded model.
- The OpenAI-style request format makes it close to a drop-in replacement in existing code, according to MakerStack.
- Perplexity states zero data retention for API traffic and publishes a SOC 2 Type II report.
Cons
- Sonar Chat Completions support ended on September 27, 2026, so existing integrations have to migrate to the Agent API, and asynchronous calls no longer work.
- Billing combines token charges with per-tool or per-request search fees, which makes costs harder to forecast than flat per-query search APIs.
- Reviewers note it gives less control over ranking and retrieval than Exa or Tavily.
- There is no ongoing free tier; a Valyu-authored comparison notes the Pro plan includes only $5 a month in API credit.
- One developer report cited by a competitor describes frequent outages while the status page showed green.
Key Features
Not included
- Custom Model Training
- Private Deployment
Perplexity Sonar Pricing
Agent API (perplexity/sonar), pay as you go
- No ongoing free tier
- $0.25 per 1M input tokens
- $2.50 per 1M output tokens
- $0.0625 per 1M cached input tokens
- Tool calls such as web search billed separately
Frequently Asked Questions
The Sonar model is still sold as perplexity/sonar in the Agent API, but Sonar Chat Completions support ended on September 27, 2026. Synchronous and streaming calls are reformulated as Agent API calls, while asynchronous calls must move to background mode.
On the Agent API, Sonar costs $0.25 per million input tokens and $2.50 per million output tokens. Tools are billed separately, for example web search at $2.50 per 1,000 invocations, or $1 with Fast Search.
No ongoing free tier. Reviewers mention only limited starter credits, and Perplexity Pro subscribers get a small monthly API credit.
Perplexity's documentation states a zero data retention policy for the API and points to a trust center with a SOC 2 Type II report and a HIPAA gap assessment.
Sonar and Sonar Pro map to the fast effort level, Sonar Reasoning Pro maps to low, and Sonar Deep Research maps to high. Perplexity recommends xhigh for the most thorough research.
No. Sonar is only available as a hosted model through Perplexity's API and third-party gateways.
Similar Tools
See all alternatives
Microsoft's free code editor with built-in GitHub Copilot chat, inline suggestions and multi-agent coding sessions, aimed at developers and engineering teams.
LangChain offers open-source frameworks (LangChain, LangGraph, Deep Agents) plus the paid LangSmith platform. Developers use them to build, evaluate, deploy and monitor LLM agents.
Pydantic AI is an open-source Python framework from the Pydantic team for building type-safe AI agents that work with many LLM providers.
Ranked near Perplexity Sonar in AI Software
See Perplexity Sonar in the listPerplexity Sonar is #393 in AI Software by our score.
