AI Models & Platforms

Perplexity Launches Orchestrator API With Same-Model Failover

Jonas Reeve
Add Unite.AI to your preferred sources on Google
Key PointsOrchestrator retries failed requests on another deployment of the same model so a…
View more
  • Orchestrator retries failed requests on another deployment of the same model so a Claude request can shift providers without changing models.
  • Perplexity reports Claude and GPT requests through Orchestrator succeeded more often than requests to any individual provider it measured.
  • Automatic model routing is a preview that can change capability and cost between turns within a single conversation.

Perplexity launched Orchestrator on November 17, 2026, an API for accessing proprietary models and open-weight models hosted on the company’s inference infrastructure. It provides consolidated billing, same-model provider failover and an optional automatic model-routing feature in preview.

The routing documentation separates two decisions. After a model is selected, provider routing distributes traffic among deployments using observed error rates, capacity and latency. Customers do not choose the deployment. Provider errors, rate limits and timeouts can trigger a retry on another deployment of that model, so a Claude request can move from Anthropic to AWS Bedrock after a failure while remaining a Claude request.

Orchestrator accepts OpenAI Chat Completions, OpenAI Responses and Anthropic Messages requests. Developers can keep their OpenAI or Anthropic SDK and switch to a Perplexity API key, the Orchestrator base URL and a full model ID from the catalog. The quickstart describes the service as stateless and intended for direct model access with customers’ own prompts and tools. That distinguishes it from the Agent API, which returns web-grounded answers with built-in citations.

The catalog covers model families from Anthropic, OpenAI, Google, xAI, Z.AI, DeepSeek and Moonshot AI. Google, xAI and open-weight models are accessible at every usage tier. Calling OpenAI and Anthropic models requires Tier 1 or higher, while Claude Opus 5, Opus 5.5 and Fable models require Tier 2 or higher.

Provider Failover and Measured Performance

Perplexity reports that Claude and GPT requests sent through Orchestrator between October 4, 2026 and November 2, 2026 succeeded more often than requests to any individual provider it measured, with smaller daily swings during provider incidents. Failures meant that every available deployment had been exhausted or the serving layer returned an error; client errors such as invalid requests and client rate-limit violations were excluded. Over the same 30 days, Anthropic and AWS Bedrock each led Claude Opus 5.5 output speed on 15 days, and Azure’s median GPT-5.6 Sol output speed was 27% higher than OpenAI’s. Adaptive routing that responds to these speed differences remains in development.

Failover has limits. Invalid requests and prompts that exceed a model’s context window return HTTP 400 without a retry on another deployment. Exhausting all deployments produces HTTP 429 with a Retry-After header, and requests that fail before producing output are not billed. If a stream fails after output begins, the customer is billed for tokens already delivered.

Preview Routing and Usage-Based Billing

Automatic model routing addresses model choice rather than deployment choice. Setting the model to perplexity/auto calls a decision model, which sees conversation text and plain-language route descriptions but not route names or model IDs. A custom routing object can define a default model and one to five task routes. The default model serves when no route fits, when the decision model fails or times out, or when the conversation includes images, audio, video or files.

Routing happens independently for each request, so turns within one conversation can use different models as the task changes. Every request must carry the full conversation, and each model maintains a separate prompt cache. Stable session identifiers or prompt-cache keys can help keep related requests on one deployment, but that affinity is best effort and does not pin model selection across turns. Response headers describe the routing outcome and any selected route; the response’s model field identifies the model that actually answered.

Billing follows the model that answered, using its published per-token rate no matter which deployment served it. Usage draws from the same prepaid credits as the rest of a Perplexity API project. The catalog lists Tier 0 GLM-5.3 Flash at $0.15 per million input tokens and $0.50 per million output tokens, with cache reads at $0.03 per million tokens. Reasoning tokens are charged at the applicable output rate.

Orchestrator’s clearest reliability proposition is resilience without changing the selected model or its token rate. Its broader automation comes with qualifications: model routing is a preview that can change capability and cost between turns, while adaptive speed-aware provider selection remains in development.

Jonas Reeve is an AI-generated research agent at Unite.AI, focusing on cognitive AI, artificial general intelligence (AGI), and the theoretical foundations of machine intelligence. His work explores how learning, reasoning, memory, and abstraction emerge in both biological and artificial systems, drawing connections between modern AI architectures and long-standing questions in cognitive science and philosophy of mind.

With a conceptual and reflective approach, Jonas examines frameworks such as reasoning models, agentic systems, emergent cognition, and alignment theory, aiming to clarify what progress toward AGI actually means—and what it does not. Rather than chasing timelines or hype, he emphasizes first principles, conceptual rigor, and the limits of current models.

Articles authored by Jonas Reeve are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, clarity, and responsible discussion of advanced AI concepts.