API models and token pricing

This public table loads currently enabled models and their token multipliers from the XEDOC API when you open this page. Refresh the list to check for changes. An authenticated /v1/models request returns the enabled models your account is allowed to use; an administrator can restrict individual accounts to a selection.

XEDOC balance tokens are prepaid account credit. The base credit price is $0.0225 per 1 million balance tokens before volume or promo bonuses. Model multipliers determine how much of that credit a request uses. The token pricing calculator shows the effective price for your selected top-up amount and currency.

Calculated request cost is uncached input × input multiplier + cached input × cached-input multiplier + output × output multiplier. Cached input is a subset of input tokens and is counted once. For a model with ×5 in every category, 1 million reported model tokens use 5 million balance tokens: $0.1125 at the base credit price before bonuses. For GPT-6 Astra API pricing, use the published ×8 row; for a lower-cost comparison, check the current model service and GPT-5.6 Luna rows. Input, cached input and output can have different multipliers.

Requests require a positive token balance and enough credit for the locally estimated text input cost, including the model multiplier. XEDOC does not reserve a full model context or impose a default output limit. Your explicit generation settings are forwarded. Complete provider usage is charged once when the request finishes. A request can deduct at most the available balance, which stops at zero. Any calculated cost above that debit is recorded separately and is not collected from future top-ups. New requests at zero are rejected and never trigger an automatic payment. If a stream ends before final usage is received, its charge needs review and does not reserve your balance.

The dashboard’s Requests view separates reported model usage from tokens actually charged. For an estimated-input rejection or exhausted balance, see HTTP 402 and token balance checks.

Use names such as claude-opus-5 exactly as shown. Claude context variants are handled by the gateway and appear once in your model list; do not add [1m] to your configuration.

Context and output limits, when reported in the API catalog, are model-service metadata. Request size and your client’s context settings can impose lower limits. A model without reported limits should not be treated as having unlimited context.

Choose the model and reasoning level in your client. The assistant provides compatible client options; the model service determines which settings a particular model accepts. Existing user context or compaction overrides can take precedence over generated catalog values.

For models without reported context metadata, generated client configuration uses a conservative context setting. That client setting is not a statement of the model’s full capacity.

Run xedoc again and select the clients you want to refresh. A paused model disappears from the current API catalog. A model saved in an older client configuration can therefore return an unavailable-model error until the list is refreshed.