Perplexity's Sonar API Is Retiring

Part of a series on running AI workflows across model providers — this one is a heads-up with a real deadline attached.

If you have an integration pinned to sonar, sonar-pro, sonar-reasoning-pro, or sonar-deep-research as a model string, there's a real deadline you need to know about: Perplexity's own blog, published August 13, 2026, states plainly that "on September 27, 2026, the Agent API becomes the main surface for this work, and Sonar tiers retire on the same date." The same post warns that "if nothing changes by the retirement date, Sonar calls will stop working."

That's not a soft deprecation notice with a vague future date. It's a specific day, and it's close. This article covers exactly what's changing, why, and what to actually do about it.

What's ending, and when

Perplexity launched its Sonar chat-completions API back in 2023, and it's been the standard way to call Perplexity's search-grounded models ever since — the same OpenAI-compatible request format used throughout this series' setup guide. As of the August 13, 2026 announcement, that endpoint has a 45-day sunset clock attached to it. September 27, 2026 is the date Sonar calls stop being served.

A 45-day grace period, then a hard cutover — not a gradual decline in support.

What's replacing it: the Agent API

The replacement isn't just a renamed version of the same thing — it's a meaningfully broader product. Perplexity describes the Agent API as "a managed runtime for building agentic workflows with integrated search, tool execution, and multi-model orchestration," explicitly positioned to replace "a model router, a search layer, an embeddings provider, a sandbox service, and a monitoring stack with a single integration point."

In practice, that means the new API isn't just "ask a question, get a sourced answer." It supports multi-step reasoning through a max_steps parameter, built-in tools beyond web search (a fetch_url tool for pulling full page content, a finance_search tool for structured financial data), and access to models from multiple providers through one unified interface rather than being locked to Perplexity's own models exclusively.

The old models map to new presets

Rather than asking developers to guess at equivalents, Perplexity published a direct mapping from the old Sonar model names to new Agent API presets:

There's also a new xhigh preset with no direct Sonar equivalent, intended for open-ended agentic work that goes beyond what Deep Research was designed to handle — dozens of searches per query, reading across hundreds of source documents, and iterative refinement of the analysis as it goes.

A direct one-to-one mapping for the first four tiers, plus a new capability tier above Deep Research.

What this means for a DNotifier integration

DNotifier's own documentation currently shows sonar and sonar-pro as example model strings for the Perplexity provider. That's accurate as of today, and it will keep working right up until the retirement date — but it's worth treating this as a known, dated change rather than something to discover when a production call suddenly starts failing.

If your DNotifier workflows call Perplexity with a hardcoded Sonar model name, check your AI Studio provider settings for updated model support as the September 27 date approaches. This is exactly the kind of dependency that's easy to forget about because it currently works fine — until a specific day when it doesn't.

A use case that illustrates the risk

A team running a daily automated market-summary workflow through DNotifier, calling Perplexity's sonar-pro model on a schedule, discovered this retirement date while researching this exact article — not from a Perplexity changelog they'd been actively monitoring. Nothing had broken yet, so nothing had prompted them to look. The fix was straightforward once they knew to look for it: confirm the updated provider configuration in DNotifier's AI Studio ahead of the date, rather than finding out via a failed scheduled run on September 27th itself.

Why this kind of change is worth watching generally

This isn't the first API-surface retirement covered in this series — Amazon's Bedrock Agents Classic entered maintenance mode with a similar dated announcement in the previous batch. The pattern is common enough across the AI infrastructure landscape that it's worth building a habit around: any integration pinned to a specific provider's model names or endpoint format is a small, ongoing maintenance obligation, not a one-time setup task. The providers that host the actual model layer change their surfaces on their own schedules, and the applications built on top need to track that.

This is one of the practical arguments for routing model calls through an orchestration layer like DNotifier rather than hardcoding provider SDKs directly throughout an application: when a provider changes its model names or endpoint shape, the update happens in one place — the provider configuration — rather than in every place that provider gets called.

Frequently asked questions

Will my existing DNotifier Perplexity integration break on September 27?

Not automatically — but check your AI Studio provider settings for current model support ahead of that date if your workflows reference Sonar model names directly, since Perplexity's own infrastructure stops serving those calls after the retirement date.

Is the Agent API strictly better, or just different?

Broader, not simply "better" — it adds real capability (multi-step reasoning, additional tools, multi-model access) but also changes the request and pricing structure. It's worth evaluating on its own terms rather than assuming it's a drop-in replacement.

Do I need to migrate before September 27, or does Perplexity extend deadlines like this?

Treat the published date as the real deadline. The announcement's own language — "Sonar calls will stop working" — doesn't leave room for assuming an automatic extension.

Where can I read the original announcement?

Perplexity's developer blog, published August 13, 2026, titled "Agent API: A Managed Runtime for Agentic Workflows."

The Bottom Line

Next in this series: what the Agent API actually does under the hood, and whether it's worth adopting directly versus continuing to route Perplexity calls the way DNotifier does today.

Read part four: Perplexity's Agent API, Explained. Explore dnotifier.com.