Skip to main content
The short answer: your tier sets your request rate, your token throughput, and the highest spending limit that you can set. Each organization begins at Base. Prem moves you to the next tier when your payments get to the amount that the tier needs.

​
Usage tiers

Prem applies tiers at the organization level. All users and API keys in an organization have the same tier. Two points about these conditions:
  • Prem counts the total amount that you paid. It does not count your current balance.
  • Prem counts the days from your first payment. It does not count the days from your most recent payment.

​
Limits by tier

The token limit applies to the chat completion endpoints. The audio limit applies to the transcription and translation endpoints. Prem measures the audio limit in seconds of audio. Both buckets refill continuously, at a rate that gives you the full limit again in one minute.
Prem applies the request limit to your organization as a whole. Prem tracks the token limit and the audio limit for each member that owns an API key. Two keys that the same member owns draw on one bucket. Two keys that different members own draw on separate buckets.
The maximum spending limit is a ceiling on the monthly limit that you set. It is not an amount that Prem bills you. It is not a cap that applies on its own. A request to set a limit above the maximum of your tier returns BUDGET_EXCEED_TIER_LIMIT.
You set your own monthly spending limit in the dashboard, up to the maximum for your tier. See Limits.

​
Request limits

Your tier sets the number of requests that you can send each second. Inference endpoints (chat completions, transcription, and translation) have their own, higher limit. Prem uses a token bucket for this limit. The burst column is the capacity of the bucket: the largest number of requests that you can send at one time. The bucket refills at the requests-per-second rate. A continuous rate above that returns a 429. Prem applies this limit to each endpoint separately. Requests to /openai/chat/completions do not consume the allowance of /rvenc/chat/completions. A request for the model list consumes neither allowance. For the mechanism and for retry code, see Rate limits.

​
Concurrent requests

Prem limits the number of inference requests that one API key can process at the same time. Your tier sets the limit: The limit applies to every inference endpoint. A request above the limit returns a 429:
The message shows the limit of your tier. The encrypted endpoint /rvenc/chat/completions always responds over text/event-stream. A call to that endpoint holds a slot until the stream closes. To run workers at the same time, supply a different API key to each worker. See Agents & Automation.

​
How to move to a higher tier

Prem examines your payment history every five minutes. Each check moves you up one tier. An organization that qualifies for Tier 3 immediately needs three checks, or approximately 15 minutes. Prem does not move you down. To see the tier of your organization, open the limits page in the dashboard. The GET /billing/tiers endpoint lists the limits of every tier, but it does not tell you which tier your organization has. If you need higher limits, contact us. The Prem team can set a custom tier or plan for your organization.

Rate limits

How Prem applies the limits, and what to do with a 429.

Limits

Set the monthly spending limit of your organization.

Models & Pricing

The price for each model, and the deployment strategies.

Production Checklist

Make sure that your capacity and cost settings are correct before you go live.