Perplexity has launched an API that lets developers access models from several AI providers through one endpoint, with failover that can switch deployments when a provider errors or times out. Its wider pitch is simpler integrations and more resilient service, though the reliability figures it cites come from Perplexity’s own measurements.
Perplexity Watch analysis
What happened
The Orchestrator API supports models from OpenAI, Anthropic, Google and xAI, alongside open-weight models including GLM, Kimi and DeepSeek. MarkTechPost reports that it accepts OpenAI Chat Completions, OpenAI Responses and Anthropic Messages requests. In many cases, teams can connect by changing their API key and base URL.
The launch also includes automatic model routing in preview. Developers can describe routes in plain language and set a default model; Perplexity says its decision model uses those descriptions to choose a route. The report details these features and the service’s terms.
Our top picks
- Same-model failover across providers
Provider errors, rate limits and timeouts can trigger a switch to another deployment of the requested model. - A familiar API surface
Support for OpenAI and Anthropic request formats may spare developers some integration work. - Plain-language routing in preview
Teams can describe routes and set a fallback model; the decision model does not see route names or model IDs. - Failed requests are not billed before output
Perplexity says requests that fail before producing a response incur no charge. - Usage stays tied to model rates
Perplexity says requests use the answering model’s published per-token rate, with no additional platform fee on top of credits.
Why it matters
A single gateway could make it easier to keep an application running through a provider outage without building and maintaining each connection separately. It also gives developers a way to direct requests to different models, although the automatic routing feature is still in preview and the report says developers cannot select a specific deployment themselves.
Perplexity says its production traffic from 4 October to 2 November had higher success rates for Claude and GPT requests than any single provider it measured, with less daily variation during incidents. It does not publish the success-rate percentages, so this is useful context, not a quantified guarantee.
Our read
This is a substantial piece of AI infrastructure, not just another model catalogue: failover and a common interface tackle the fiddly work between a model and the app relying on it. The practical test will be whether the promised reliability and routing control hold up across workloads, and whether the credit-purchase tiers suit teams that want access to the most expensive models.
What to watch
- Whether Perplexity publishes comparable success-rate figures and methodology.
- How automatic routing performs beyond its preview stage.
- Whether deployment choice and other routing controls expand.
- How the credit-purchase thresholds affect access to higher-tier models.
Discussion spark: Would you trust a single API gateway to choose where your model request runs, or is control over the serving provider worth the extra integration work?
Sources and evidence
- TOP 7 Platforms for Improving Brand Visibility in ChatGPT and Perplexity – Root-Nation.com (19 November 2026, 21:03 UTC)
- Perplexity Launches Orchestrator API for Failover and Model Routing – MarkTechPost (19 November 2026, 21:03 UTC)
Independent WittyWires tracker for public updates about Perplexity. Not affiliated with or endorsed by Perplexity; this is not an official account.