Starter
two tiersFor getting an agent or a product onto RouterLane.
- Fast and Normal
- 60 requests a minute
- 4 concurrent requests
- 20M tokens a month
Pricing
Pay for the lane.Not for a subscription.
One price per tier, per million tokens, whichever model answers. Metered per request and billed monthly. No subscription fee.
Rates per tier
| Tier | Model ID | Input | Output | Cached input |
|---|---|---|---|---|
| Fast | fast | $0.15 | $0.60 | $0.03 |
| Normal | normal | $0.40 | $1.60 | $0.08 |
| Smart | smart | $1.00 | $4.00 | $0.20 |
| Smart+ | smart-plus | $1.00 | $4.00 | $0.20 |
USD per 1M tokens. Cached input is input the model reads from its prompt cache. Smart+ bills every token its agents spend at Smart rates.
Three plans.Same prices on all of them.
A plan sets which tiers you can call and how fast. The per-token prices above apply on every plan.
Pro
adds SmartFor teams that run agents all day.
- Fast, Normal and Smart
- 300 requests a minute
- 16 concurrent requests
- No monthly token cap
Enterprise
adds Smart+A dedicated deployment with its own console.
- All four tiers, including Smart+
- Limits set to your traffic
- Transcript explorer, ask the data, session summaries
- Content filters and the retention you set
Estimate a month.Before it happens.
Enter millions of tokens per month per tier. Agent loops resend their history every turn, so a large share of their input is usually served from the cache.
| Tier | Input, M tokens | Output, M tokens | Cost |
|---|---|---|---|
| Fast | $2.76 | ||
| Normal | $14.72 | ||
| Smart | $9.20 | ||
| Smart+ | $0.00 |
Estimated monthly bill
$26.68
77M tokens
- Plan that fits
- Pro
Smart is included on Pro and Enterprise.
What you never pay for.Checked against the meter.
Cache hits
An identical request repeated within 10 minutes is answered from the response cache, free. The response carries x-router-cache: hit.
The decision
Reading the task and choosing the model is on us, on every new task.
Failed attempts
When a model fails before the first byte and the router moves to the next one, only the attempt that answered is metered.
Requests no model could answer
If every attempt fails, the error comes back to you and nothing is billed.
Refused requests
A bad key, a tier outside your plan, a limit reached, or a filter that blocks the input: the request is never forwarded and never billed.
Smart+ uploads
The repository goes straight to RouterLane over HTTPS, never through a model, and costs no tokens.
Billing questions
When am I billed?
Every request is metered as it happens, with its input, output and cached tokens. You are billed monthly for what you used. There is no subscription fee.
Does the price change when the router uses a bigger model?
No. You pay your tier's rate whichever model answers. When Normal sends a hard task to GLM 5.3, it is still billed at Normal's rate.
What counts as cached input?
Input the model reads from its prompt cache instead of processing again. It is billed at the cached rate, a fifth of the input rate. The router keeps each conversation on one model so that cache stays warm.
How is Smart+ billed?
Every token the job's agents spend is billed at Smart rates, on the request for the turn that started the job. Nothing is billed twice. Each job has a spend ceiling, $1.50 by default: a job that reaches it stops, delivers what it has, and says so.
What happens when I reach a limit?
Requests past your plan's rate, concurrency, monthly token or monthly spend limit get a 429 with a message that names the limit. Monthly limits reset at the start of each calendar month, UTC.
Can I cap my spend?
Yes. Ask for a monthly spend cap on your account and requests past it are refused with a 429 until the month turns. On Enterprise you set caps per plan and per client in the console.
Can I see what a single request cost?
Every response carries an x-router-request-id. Quote it to support for the request's tokens, model and cost. Enterprise consoles show cost on every request, turn and session.
Pick a lane.Ship the work.
Keys are issued on request. Tell us what you are building and which clients you use, and we will send a key by email.