Custommodel combos with auto failover
Create a virtual model name bound to multiple real models and auto-schedule by primary/backup, round-robin or random. When a model is unavailable, it switches to the next automatically
Create your smart route in 3 steps
Create a route on the Console's "Smart Routing" page
Pick a strategy (primary/backup, round-robin, random), add multiple models, set priorities
{身份码}/{路由名}
model: abc123/k2w8xThe system auto-schedules by strategy and switches on failure
Three scheduling strategies
Pick the best scheduling approach for your scenario
main_backup
Tries models in priority order. The primary model is preferred; on failure it auto-switches to a backup. Great for scenarios requiring high stability.
round_robin
Each request rotates to the next model, evenly distributing load. Great for spreading request volume and avoiding overloading a single model.
random
Each request picks a model at random — simple and efficient. Great for multiple equal-quality models with no specific priority.
How it works
Strategy scheduling → model attempts → failover → return result
Parse route name
Recognizes the {identity}/{route_name} format and looks up the user's route config and its bound model list
Strategy scheduling
Generates a model attempt queue based on the strategy (primary/backup, round-robin, random), sorted by priority
Try in turn
Calls models in queue order; on failure auto-switches to the next, supporting channel-level failover
Return result
The first successful model's result is returned to the user — fully transparent, same experience as a regular model
Use cases
When should you use smart routing?
When the primary model is unstable, auto-switch to a backup to ensure business continuity. Great for critical calls in production that can't tolerate failure.
Rotate across equal-quality models to spread request volume and avoid hitting a single model's rate limits or quota exhaustion.
Prefer cheap lightweight models and fall back to expensive-but-stable ones on failure. Balance cost and quality.
Give your team or app a fixed virtual model name; backend model changes don't affect frontend code. Just edit the route config to switch models.
Auto failover
Auto-switches to the next model on failure
Channel-level switching
Multi-channel failover within the same model
Flexible strategies
Pick from primary/backup, round-robin or random
Transparent
Same experience as a regular model
Billing notes
Each model billed independently
The actually-called model is billed at its route's normal rate. Plan users deduct plan calls. Failed models are not charged.
One log per request
No matter how many models were tried, only one call log is produced. The switching process is recorded in the failover details.
FAQ
What's the difference between smart routing and multi-model fusion?
Smart routing tries models in turn (returns on the first success); multi-model fusion calls all models in parallel and a judge fuses the results. Smart routing pursues high availability; fusion pursues high quality.
What's the identity code in the call name?
The identity code is your account's unique identifier (User.InviteCode). You can copy the full call name from the Console's "Smart Routing" page.
Is streaming supported?
Fully supported. Both streaming and non-streaming work, and the switching is transparent to the client. Cherry Studio, Cursor and other tools work normally.
Can I pick any model?
You can freely pick from all platform routes, the free route, and models included in your plan. See the Console's "Smart Routing" page for the list of available models.
Does switching affect speed?
Only when the primary model fails does it try the next; normally there's no extra latency. The first request establishes channel stickiness, and subsequent requests prefer the channel that last succeeded.
Create your first smart route
Go to the Console's "Smart Routing" page to configure a custom model combo
Go create