Hot AI News

Perplexity’s Orchestrator API puts 85+ models behind one key and no platform fee

Perplexity opens the routing layer it runs in production to developers: one API key, failover across providers, unified billing and no platform fee on top of credits.

Perplexity’s Orchestrator API puts 85+ models behind one key and no platform fee
Image: Perplexity

Perplexity has launched the Orchestrator API, a single endpoint that lets developers send requests to models from OpenAI, Anthropic, Google, xAI and several open-weight families, with provider failover and one invoice. According to the official announcement, Perplexity adds no platform fee on top of credits.

One endpoint, familiar SDKs, three API formats

Orchestrator covers models from OpenAI, Anthropic, Google and xAI, plus open-weight models such as GLM, Kimi and DeepSeek. Perplexity says open-weight models are served either by the company itself or by inference partners. The API accepts the OpenAI Chat Completions, OpenAI Responses and Anthropic Messages formats, so teams can keep their current SDKs. For most, Perplexity says, getting started means swapping the API key and base URL and picking a model from the catalog.

Perplexity co-founder and CEO Aravind Srinivas, wearing a Perplexity sweatshirt, speaks onstage at TechCrunch Disrupt 2024.
Photo: TechCrunch / Wikimedia Commons, CC BY 2.0

Teams choose which model handles each workload. Perplexity takes care of provider routing and failover. The company describes the pitch as running simple, high-volume tasks on cheaper models while saving stronger ones for harder jobs.

Billing is consolidated. Perplexity argues that using several labs and inference hosts normally means separate accounts, contracts, security reviews and annual commitments tied to a model mix that can change within weeks. Orchestrator puts supported models under one vendor and one bill. The API Console shows spend by model, project, API key and date, and offers admin controls: model allowlists for an organization, user restrictions, spend caps and capacity requests.

Advertisements

Failover that Perplexity says beats single providers

Orchestrator tracks health, capacity and latency across providers and spreads traffic toward the best performers. If a request fails, it can be retried on another deployment of the same model. Perplexity says the model layer behind this already handles hundreds of millions of requests across more than 85 models and over a dozen providers.

The company makes one measurable claim. Over a month of live Orchestrator traffic, it says, its Claude and GPT calls succeeded more often than they did on any single provider it tracked, and varied less from day to day when providers had outages. No figures, provider names or methodology are published in the post, and the claim has not been independently verified.

A gap between 99.0% and 99.99% success can affect thousands of responses for teams running millions of requests a day.

Two features are in preview. Routes let teams describe in plain language which requests should go to which model, with a default model as fallback and built-in Lower cost, Balanced and Higher quality profiles as starting points. Auto effort changes how much reasoning a supported model applies on its next turn, following a policy the developer sets, so an agent can try harder after a tool error without changing models.

Who benefits and what is still unclear

The practical audience is teams that already mix models and are tired of managing the contracts. Prices in this market move fast: Google made Gemini 4 Carbon generally available on November 10 at $2 per million input tokens and $10 per million output tokens, according to our earlier report, with those rates described as introductory through November 30. A gateway that lets a team move traffic to a cheaper or newer model without a new contract is aimed squarely at that churn.

It also extends a pattern in Perplexity’s developer platform. Its Monitor API, covered in our report, is priced at $1 per 1,000 successful runs and shares an API key with the company’s Search, Context and Agent APIs. Orchestrator adds the model-access layer to that same account. On the consumer side, Multiagent already shows Perplexity coordinating several frontier models inside one task.

The announcement leaves several things open. It does not publish per-model prices or say whether they match the labs’ list prices, only that no platform fee is added on top of credits. Rate limits, latency overhead from routing, regional availability and data-handling terms are not covered. Teams with large or committed workloads are told to email [email protected] for volume pricing and capacity. Perplexity also links separate material on how it built and evaluated Routes, which was not part of the page quoted here.

For developers, the test will be whether the reliability numbers hold up outside Perplexity’s own traffic, and whether credits cost the same as going direct. Both are measurable, and neither is answered yet.

Sign Up for the Daily AI Pulse

One email a day. All the stories that matter.

Alex CarterAI News & Big Tech Correspondent

Alex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may also like