Models
Which providers Operant speaks to, and how a call gets assigned to a model.
Providers
| Provider | Status |
|---|---|
| Anthropic | Supported by default |
| OpenAI | Supported by default |
| Anything behind a LiteLLM-compatible proxy | Opt-in, configured by you |
You bring the credentials. Operant does not resell inference, and your provider invoices you directly.
Why there is no price list on this page
Operant carries an internal table of per-model prices, and it changes whenever a provider changes theirs. Publishing a snapshot of it would produce a page that is quietly wrong within weeks, and a routing decision made against a stale price is worse than no page at all. The prices the Ledger shows you are the ones routing actually used.
How a model is chosen
Per call, by task shape rather than by a static mapping — with the KV cache accounted for, because switching models discards it and the discard sometimes costs more than the cheaper model saves. The Gateway describes the mechanism.
You can always pin a workload to an exact model. Routing is a default, not a lock-in.