Perplexity launches API for provider failover and AI model routing

Executive Briefing
“Perplexity has launched Orchestrator API, giving applications one endpoint for multiple AI...”
Key Takeaways
- 1.According to the reported launch details, the API supports models from OpenAI, Anthropic, Google and xAI, alongside open-weight models including GLM, Kimi and DeepSeek.
- 2.Failover keeps the requested model but changes its deployment after a provider error, rate limit or timeout. A Claude request, for example, could move from Anthropic to AWS Bedrock.
- 3.Automatic model routing is in preview. Developers can define up to five routes, with a default model answering if none matches or the routing decision fails.
- 4.Perplexity reported that, in traffic measured from October 4 to November 2, 2026, Claude and GPT requests through its service succeeded more often than requests to any individual provider it measured. The article gives no success-rate percentages.
- 5.Under the stated pricing policy, requests are billed at the answering model’s published token rate. Provider failover does not change that rate, and Perplexity says it adds no platform fee on top of credits.
Perplexity has launched Orchestrator API, giving applications one endpoint for multiple AI models. It can switch providers when a deployment fails, while automatic model routing remains in preview.
Product Features & Key Highlights
According to the reported launch details, the API supports models from OpenAI, Anthropic, Google and xAI, alongside open-weight models including GLM, Kimi and DeepSeek.
Failover keeps the requested model but changes its deployment after a provider error, rate limit or timeout. A Claude request, for example, could move from Anthropic to AWS Bedrock.
Automatic model routing is in preview. Developers can define up to five routes, with a default model answering if none matches or the routing decision fails.
Perplexity reported that, in traffic measured from October 4 to November 2, 2026, Claude and GPT requests through its service succeeded more often than requests to any individual provider it measured. The article gives no success-rate percentages.
Under the stated pricing policy, requests are billed at the answering model’s published token rate. Provider failover does not change that rate, and Perplexity says it adds no platform fee on top of credits.
Practical Assessment & Evaluation
The two mechanisms address different problems. Failover can preserve the model when one deployment has trouble; automatic routing can send different tasks to different models. Applications would need to test answer quality for each route before broad use.
One endpoint and consolidated billing could reduce integration work. Perplexity’s reported measurements, however, do not establish how much reliability would improve for a particular application.
Pricing, Availability & Next Steps
The source does not provide specific success rates or the timeout thresholds that govern failover.
It remains unclear how consistently automatic routing will perform, or what it will cost, across workloads beyond the described cases.







