Auto Routers Explained
Let Krater pick the best model for every message Auto Routers remove the guesswork: choose a cost tier and the router sends your message to a model selected for that tier. Every reply shows which m...
Let Krater pick the best model for every message
Auto Routers remove the guesswork: choose a cost tier and the router sends your message to a model selected for that tier. Every reply shows which model answered.
There are five Auto Routers on Ultra and Max plans. The picker shows $ for Lite, $$ for Smart, and $$$ for Power, Elite, and Frontier:
| Router | What it picks | Picker tier |
|---|---|---|
| Lite Auto Router | Krater picks the most cost-efficient model for the job | $ |
| Smart Auto Router | Krater picks the best model for every prompt | $$ |
| Power Auto Router | Krater picks a stronger model for harder work | $$$ |
| Elite Auto Router | Krater picks a top-tier model for demanding tasks | $$$ |
| Frontier Auto Router | Krater picks the most capable models, regardless of price | $$$ |
On Ultra and Max, Smart is the default model for new chats, so routed answers are the out-of-the-box behavior.
Using a router
- Click the model selector in the chat input.
- Pick a router. Routers appear inline in the All Models list, sorted by real usage, or flip the Routers pill to show only routers.
- Send your message. The router picks a model and answers with it.
To see which model answered, hover the model badge on the reply. It shows the actual model plus which router selected it, for example “Selected by Smart Auto Router”. You can switch routers (or jump to a concrete model) mid-conversation; context is preserved.
How routing works
- Each router targets a cost tier and picks a model for that tier from the full catalog, with no fixed model list.
- For a specific model or version, choose it directly.
- Under private-only enforcement, any Auto Router is limited to first-party providers: OpenAI, Anthropic, Google, Mistral, Cohere, and xAI.
- Routers are chat-only. They cannot be used in /compare battles, which run on concrete models.
Cost
- You are billed at the picked model’s real cost. Lite costs least, and Frontier costs most.
- The picker uses relative cost tiers: $ for Lite, $$ for Smart, and $$$ for Power, Elite, and Frontier. These are not fixed credit prices.
- The cost of each reply depends on the selected model and its usage.
Who can use Auto Routers
| Plan | Access |
|---|---|
| Ultra | All 5 routers; Smart is the default model |
| Max | All 5 routers; Smart is the default model |
| Pro | Not included. Upgrade to unlock |
| Personas | Persona chats always run the Smart router, on every plan including Pro |
- Selecting a router on a plan without access returns
AUTO_ROUTER_PLAN_REQUIRED. See error codes for the fix. - Accounts with Only use Private models enabled (Settings > Data & Privacy, Max plan) or a team-enforced private policy can use any Auto Router. Its model choice is limited to first-party providers: OpenAI, Anthropic, Google, Mistral, Cohere, and xAI.
Auto Routers in the API
Use auto/lite, auto/smart, auto/power, auto/elite, or auto/frontier as the model value on the chat endpoint. Older router IDs remain supported and map to the closest current router. API access requires Ultra or Max with an API key from Settings > API.
Tips
- Not sure? Start with Smart, the balanced default.
- Two-pass work: draft with Lite, then use Power for a harder final pass.
- Private-only policy: any Auto Router remains available and can select only first-party providers.
- Check the model badge on each reply to see which model the router selected.
Related: Chat models / Understanding credits / Plans overview / Compare models