NewsStocksPerplexity Launches Orchestrator API Unifying Access to Hosted Open-Weight Models

Perplexity Launches Orchestrator API Unifying Access to Hosted Open-Weight Models

Author: CryptoBriefing·

Key Takeaways

  • •Perplexity's Orchestrator API entered private preview on November 9, 2026, giving developers a single endpoint that selects models and providers for a given task and merges costs into one bill.
  • •The API accepts both OpenAI and Anthropic request schemas, allowing any supported schema to call any hosted model regardless of which lab built it.
  • •The initial catalog includes open-weight models from Moonshot AI, Z.AI, and DeepSeek, with pricing published per million tokens and a models endpoint for querying the full list.
  • •Each model ID ships with automatic request routing, health monitoring, and failover, while billing is pay-as-you-go and rejected requests carry no charge.
  • •Access currently requires emailing for the private preview, and the launch follows the Agent API as Perplexity reports an ARR run rate exceeding $1 billion and considers accelerating an IPO toward 2027.
Perplexity Launches Orchestrator API Unifying Access to Hosted Open-Weight Models

Perplexity has launched its Orchestrator API, a single gateway that selects models and providers for a given task and consolidates the associated costs into one bill. The service went live in private preview on November 9, 2026, and is aimed squarely at developers who are tired of maintaining a separate integration for every model they want to try.

What the Orchestrator API Actually Does

The API gives developers unified access to a catalog of open-weight models that Perplexity hosts itself. Open-weight means the underlying model parameters are publicly released, which allows a company like Perplexity to run them on its own infrastructure — so individual teams do not have to stand up their own serving stack for each model.

Everything runs through a single base URL: ` Developers authenticate with their existing Perplexity API key, meaning there is no new credential to manage.

The compatibility layer is the notable part. The API accepts request formats from both OpenAI and Anthropic, the two schemas most developers already have baked into their code. Perplexity says developers can use any supported schema regardless of which lab originally built the model. That decoupling is the core pitch: the format a developer writes in no longer dictates the model they can call. For teams juggling multi-model setups, that folds what would otherwise be a patchwork of provider-specific endpoints and separate invoices into one integration and one bill.

The Model Lineup and the Reliability Layer

The initial catalog leans on prominent open-weight releases:

  • perplexity/kimi-k3, from Moonshot AI
  • perplexity/glm-5.3 and perplexity/glm-5.3-flash, from Z.AI
  • perplexity/deepseek-v4-pro-0813, from DeepSeek

Developers can pull the full list programmatically through a GET /orchestrator/v1/models call. Pricing is published per million tokens for each model, making side-by-side cost comparisons across the catalog straightforward.

Each model ID comes with automatic request routing, health monitoring, and failover across all deployments. If one deployment stumbles, traffic is shifted elsewhere rather than returning an error to the user — callers keep hitting the same endpoint even as requests are rerouted behind the scenes.

How Billing Works

Billing is pay-as-you-go, based on token usage at the published per-token rates. Tokens are the small chunks of text that models read and write, so the bill scales with how much work the model actually does. Rejected requests, such as those that hit rate-limit errors, carry no charge.

Access is not open to everyone yet. Developers who want in during the private preview need to email [email protected].

Background: A Busy 2026 for Perplexity

The Orchestrator API is the latest in a string of developer-focused releases from Perplexity this year. Earlier in 2026, the company introduced its Agent API, and the new launch builds on that push into enterprise tooling. The timing also lines up with other November 2026 initiatives, including Multiagent effort levels and Frontier Research projects.

The business numbers suggest the strategy is landing. Perplexity reported an annual recurring revenue run rate exceeding $1 billion as of mid-November 2026, a figure described as significantly higher than the prior year. Company leadership has also floated the possibility of accelerating toward an IPO by 2027 as revenue climbs. Read against that cadence, the gateway looks like a continuation of the developer-platform push rather than a one-off. From here, the details worth tracking are whether access widens beyond the email-gated preview and what lands next in the hosted catalog, which developers can query at any time through the models endpoint.

Disclosure: This article is based on reporting from CryptoBriefing, edited by Diego Almada Lopez.