Perplexity launches Orchestrator API with model routing and same-model provider failover
Orchestrator API puts open-weight and proprietary models behind one endpoint with unified billing and automatic failover between providers serving the same model. Optional automatic model routing is available in preview.

AT A GLANCE
Key points
- One API combines open-weight and proprietary models with unified billing.
- The same model can be retried on another deployment after a provider failure.
- Task-based automatic model routing is available in preview.
What is changing?
Perplexity launched Orchestrator API on November 17. The service brings models from OpenAI, Anthropic, Google and xAI together with open-weight models such as GLM, Kimi and DeepSeek under one API. It accepts OpenAI Chat Completions, OpenAI Responses and Anthropic Messages formats, so an existing SDK can be connected primarily by changing the API key and base URL.
Orchestrator provides two distinct routing layers. Provider routing watches health, capacity and latency across deployments of the selected model; after an error, rate limit or timeout, it can retry that same model through another provider. A Claude request, for example, can be served through AWS Bedrock instead of Anthropic without changing the selected model. Perplexity says provider failover does not change the model’s published token rate and usage draws from prepaid API credits.
Automatic routing, currently in preview, sends requests to different models according to tasks described in plain language. Teams can define their own profile or use preset cost, balanced and quality profiles. Perplexity’s measurements from October 4 to November 2 claim that Claude and GPT requests on its infrastructure achieved higher success rates than the individual providers it measured; this provider-run result does not guarantee latency or accuracy in every application. Orchestrator may require separate access approval, and volume pricing is not publicly specified.
This report is prepared with AI assistance and checked against sources. Automated checks do not constitute human editorial review. Editorial approach
Explore the original sources
Source date: 17 November 2026.
Published by AI Radar: 18 Nov 2026, 00:07 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.






Loading comments…