Models and pricing

The runtime source of truth is the authenticated model catalog. For an OpenAI-compatible client, call GET /v1/models with an API-credit key or GET /token/v1/models with an API-token key. It lists only models with active pricing and a configured route.

List live models

GET $CLOUDSERVICE_BASE_URL/models

Set CLOUDSERVICE_BASE_URL to https://api.cloudservice.services/v1 for API credit or https://api.cloudservice.services/token/v1 for API tokens. Native Anthropic clients discover models through their origin plus /v1/models.

curl "$CLOUDSERVICE_BASE_URL/models" \
  -H "Authorization: Bearer YOUR_CLOUDSERVICE_API_KEY"

Pricing

Every model has a published customer-facing input rate and output rate, plus a cached-read input rate where the upstream provider supports one. Uncached input and cache writes bill at the model’s input rate. Provider-reported cached-read input bills at the model’s cached rate where one is published; on models with no published cached rate, cached input bills at that model’s ordinary input rate instead. A published cached rate covers provider-reported cached-read input only and does not change with context length. The ledger split behind these rules is described in Pricing and accounting architecture, and pack prices are listed in Billing, API credits & token packs.

GPT-5.6 long-context pricing

These models have a 1,050,000-token context window and a 128,000-token maximum output. When a request contains more than 272K input tokens, long-context rates apply to the entire request—not only the tokens above the threshold. Above the threshold, uncached input and output move to the long-context rates; a published cached-input rate is flat and does not change with context length. Internal provider cache prices remain separate from customer billing. Fable does not have a long-context tier.

Choose a model

Use the exact IDs returned by the API. Capability details can vary by provider route, so test the model in your workload before a large rollout.