Introducing Orchestrator: One API for frontier and open-weight models

Perplexity Orchestrator provides access to frontier and open-weight models with unified billing and failover in one API.

An orchestra conductor surrounded by sweeping trails of light against a dark background.

Perplexity is multi-model at its core, giving people the freedom to choose the latest frontier and open-weight models for their work. This multi-model approach draws on each model’s strengths and can lower costs when a task doesn’t need the most powerful model. Behind the scenes, our orchestration technology can use several models to complete a single job.

Businesses are taking a similar approach to balance performance, cost, and flexibility. Each new model release presents an opportunity to lower costs or improve quality. This year alone, labs such as Anthropic and OpenAI have released more than a dozen new models. More than 100 open-weight models have also shipped. While this gives teams more options for each workload, it also gives them more providers, contracts, and invoices to manage.

Today we’re launching the Perplexity Orchestrator API, making our model-agnostic technology available to developers and teams building AI applications. Through one API, teams can run simple, high-volume tasks on lower-cost models and reserve more powerful models for more demanding work.

With Orchestrator, teams can lower and centralize AI spend and adopt new models as they ship, without compromising on performance or uptime. Teams can choose which model handles each workload, and Perplexity handles provider routing and failover to ensure reliable inference.

Built on Perplexity's production model layer

Perplexity runs open-weight and frontier models in production every day. Our model layer routes hundreds of millions of requests across 85+ models and more than a dozen providers.

Choosing a model is easier than keeping every request fast and affordable when providers slow down, run short on capacity, or change prices. We’ve lowered our own serving costs by moving workloads to models best suited to each task.

Orchestrator gives developers access to the shared infrastructure behind that approach.

Orchestrator makes that inference layer available through one API. It covers models from OpenAI, Anthropic, Google, and xAI, alongside open-weight models such as GLM, Kimi, and DeepSeek. Open-weight models are served by Perplexity or through inference partners.

Orchestrator supports the OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages formats. Teams can continue using familiar software development kits (SDKs). For most teams, getting started involves updating the API key and base URL, and selecting an available model.

Unified billing with no additional platform fee

Running models from several labs and inference providers can require separate accounts, API keys, contracts, security reviews, and invoices. It can also mean separate annual commitments based on a model mix that could change within weeks.

Orchestrator brings supported models under one vendor and one bill, so teams can change their model mix without negotiating a separate commitment with each provider. There’s no additional platform fee on top of credits.

Teams can also track usage and manage spending through the API Console:

  • Usage reporting: Spend broken down by model, project, API key, and date
  • API keys: Separate API keys to track usage by project or team
  • Admin controls: Organization model allowlists, user restrictions, spend caps, and capacity requests

More reliable requests across providers

Models available through multiple providers give teams more than one way to serve a request.

Orchestrator tracks the health, capacity, and latency of each provider to distribute traffic across the best-performing ones. If a request fails, it can retry on another deployment of the same model, handling failover without requiring teams to build the logic themselves.

For teams running millions of requests a day, even a small difference in request success rates, a delta between 99.0% and 99.99%, can affect thousands of responses. With Orchestrator, teams don't have to build or maintain failover code themselves.

In one month’s production traffic on Orchestrator, Claude and GPT requests through Perplexity had higher success rates than any individual provider measured, along with less daily variation during provider incidents.

In preview: Routes and Auto effort

Routes let teams describe in plain language which requests each model should handle, and set a default model. Teams can start from a built-in Lower cost, Balanced, or Higher quality profile. Or they can build a custom one. If teams don’t provide a description or specify a goal, the default model serves the request.

Also in preview, Auto effort adjusts how much reasoning a supported model uses on its next turn according to a policy the developer defines. An agent can spend more effort recovering from a tool error without switching models.

Get started with intelligent model orchestration

Orchestrator is the first step within Perplexity's API platform for intelligent model orchestration. Learn how we built and evaluated Routes, including why model routing is hard to get right.

To get started, create an API key in the API Console, point an existing OpenAI or Anthropic SDK at the Orchestrator base URL, and choose from the model catalog.

The quickstart covers a first request in minutes. Teams with large or committed workloads can email api@perplexity.ai for more information on volume pricing and capacity.

Create an API key today.