Usage and quotas

Which models you can use, your rolling 30-day usage allowance for self-service LLM keys, the 429 you get over the cap, and optional hep_tokens overage.

Usage and quotas#

Self-service LLM keys are metered against a rolling dollar-value allowance. This page explains which models you can use, how much usage is included, and what happens when you run past it.

Models and plan access#

GET https://ai.hep.gg/models lists the slugs you can mint a key for. If you pick a model that is not available on your plan, key creation returns 403.

Your monthly allowance#

Usage is valued in dollars at each model's rate and charged against a rolling 30-day allowance for your whole account (every key you own counts against the same pool):

"Rolling" means the window is always the last 30 days, so old usage ages off continuously and your headroom refills gradually. It is not a hard reset on a fixed day of the month. The qwen3-8b rate is $0.06 per 1M input tokens and $0.24 per 1M output tokens, metered separately. Your current spend and remaining allowance are shown on the LLM Keys page.

When you hit the cap#

Once your rolling usage reaches your allowance, POST /v1/chat/completions returns 429 in the OpenAI error shape with type: "quota_exceeded":

{
  "error": {
    "message": "Monthly usage limit reached ($10). It refills as your last-30-day usage ages off, or enable extra usage (hep_tokens) in your dashboard.",
    "type": "quota_exceeded",
    "code": "quota_exhausted"
  }
}

The code tells you why:

Extra usage (hep_tokens overage)#

Extra usage is off by default. Turn it on from the LLM Keys page to keep working past your included allowance. Beyond the allowance, each request is billed to your hep_tokens balance at the standard list rate of 240 hep_tokens per $1 of usage.

• You can set an optional cap (the most hep_tokens to auto-spend per rolling 30 days). 0 means no cap, up to your balance.
• When your balance reaches 0 or you hit your cap, requests return 429 again (no_hep_tokens or overage_cap).

Admins#

Admin accounts have no quota, can use every model, and keep the existing admin key path. None of the limits above apply to them.