Help
Glossary
One-line definitions of the terms used across the docs.
The words the docs and the API use, in one line each.
Terms
| Term | Definition |
|---|---|
| Slug | A model's one id, such as claude-sonnet-5.5. Other spellings resolve to it. |
| Spelling | An alternative id for a slug: <vendor>/<slug>, a dashed Anthropic id, or <slug>:free. |
| Catalog | Every model you can call, with prices, context and providers. Listed equals callable. |
| Waterfall | A model's ordered list of rungs. |
| Rung (route, lane) | One way to reach a model: a provider and its model id. |
| Failover | Moving a request to the next rung after a provider failure, before any output. |
| Route depth | Position of the serving rung; 0 is the first. |
| Platform-funded | A rung paid with your credits at catalog price. |
| Pass-through (BYOK) | A rung on your own provider key; your provider bills you. |
| BYOK | Bring your own key: connecting your own provider account. |
| Experiential Cloud | Serving capacity Experiential manages, shown as provider experiential_cloud. |
| Credit | One cent of platform usage. |
| Nano-USD | The API's money unit: 1,000,000,000 nano-USD is $1. |
| Hold | A reserved worst-case cost while a request runs; it settles to the actual cost. |
| Card verification | A temporary $1 hold that unlocks free and promotional models. |
| Plan discount | A percent off list on eligible models for Pro, Max and Ultra. |
| Daily included allowance | The credits per UTC day a plan's discounts and plan-free models can use. |
| Rate-limit tier | Your organization's requests and tokens per window, set by what you bought. |
| Free core | The free allowance on a few models for verified organizations. |
| Promotion | A time-bound free or discounted offer on a model, with token limits. |
| Cache read | Input tokens served from a provider's prompt cache, at the cached rate. |
| Cache write | Input tokens stored into a provider's prompt cache, at the write rate. |
| Prompt capture | The organization switch that stores prompts and responses. |
| ZDR | Zero data retention: the provider keeps no request content. |
| ZDR on request | A route that keeps no content when the request carries a zero-retention constraint. |
| No-training | Routing only to providers that do not train on API data. |
| Provider policy | Your organization's rules for which providers requests may reach. |
| Identity | A group of keys (a team, a product) for usage, budgets and guardrails. |
| Provisioning key | A key that can also create and revoke other keys. |
| Guardrails | Content checks you configure per identity on requests and completions. |
safety_identifier | Your end customer's opaque id, used by resellers for attribution. |
| Uncensored model | A model with reduced refusals, subject to content moderation. |
| See also | |
|---|---|
| FAQ | Short answers |
| Routing | Waterfalls and failover |
| Data controls | ZDR and capture |
| Plans and pricing | Tiers and discounts |