Introducing the Perplexity Orchestrator API.
@perplexity_ai514K followers
266,203 views
25 Comments
Comments from the original post on X (Twitter)

@perplexity_ai
Orchestrator API serves 45+ models from OpenAI, Anthropic, Google, xAI, as well as open-weight models. You can connect an existing OpenAI or Anthropic SDK to a Perplexity API key and all models can be called with OpenAI Chat Completions, Responses, and Anthropic Messages.
8 likes

@perplexity_ai
Orchestrator offers automatic routing, available in preview. It picks a model per request using profiles that map plain language tasks to models, then uses a decision model to match the conversation to the right route.
7 likes

@perplexity_ai
Orchestrator is built on Perplexity's model layer, which runs in production every day, intelligently routing hundreds of millions of requests across 85+ models and more than a dozen providers. Now developers can build on the same infrastructure.
6 likes

@perplexity_ai
Each model is billed at its published per-token rate, whichever provider serves it. Track spend by model, project, and API key in the API Console, and set model allowlists and spend caps. Read more:
8 likes

@biscuitweb
rerouting refusals to another model will get some angry emails from safety teams
1 like

@mhdfaran
unified billing is the underrated part, one invoice beats juggling five API accounts.
1 like

@jeffbruchado
if the provider can change per request, every eval run has to log which one actually served each call or the scores stop being comparable

@jason_coder0
Smarter model routing and unified API access. A major upgrade for developers.

@lozoboss
Does it tell you which model it actually routed to on each call? That's the first thing I'd want logged

@AIQuanting
Say it routes to a pricier host for the same open model because latency was better. Does the bill show that, or is it one flat price per model?

@heyartzy
Unified billing for model routing is a solid move

@bobolv235824
"picks the right model automatically" means you're trusting the router's incentives, not the model's. cheapest-that-passes is a very different goal from best-for-you.

@Lmin0711
When it auto-routes per request, do responses show which model actually ran? And can you set a fallback order, or is it fully automatic?

@smobilefactory
does it fall back on its own if a provider dies mid request?

@AnandButani
routing is a solved-ish problem. unified billing isn't. cached vs uncached token rates alone differ per provider, and invoices never reconcile cleanly

@swordBluesy
let the router pick, but i pin goal, scope, acceptance criteria in the profile so it doesn't hallucinate. we hit this making 好汉重头来.

@Shoier__
Why did you remove deep search from normal chat and made it in Computer? This is so annoying

@ageshah
Great idea, but isn't one of the big cost savings input cache? Wouldn't we lose the benefit of that with dynamic routing per call?

@CalebBaskin
Do Enterprise Max subscribers get any API allocation with our subs yet?

@raulnogales
One API with unified billing is the part teams will feel first. Choosing the right model per task only helps if the cost and the provider switching stay invisible behind it.




