Concepts and glossary
Tokens
Section titled “Tokens”A token is the unit the API counts. A chat request can set max_ to cap how many tokens the model generates. Rate limits also count tokens per minute.
TPM means tokens per minute. Each shared plan publishes an input TPM and an output TPM for every model. Those figures, and the conditions on them, are on Rate limits.
RPM means requests per minute. Each shared plan publishes how many requests you can send to a model. RPM is published on the same Rate limits page as TPM.
Context window
Section titled “Context window”The context window is how much text a model accepts in one request. The catalog lists it for each model, for example 8K, 128K, or 1M. Recommended models shows the value next to each recommendation.
T-Cloud
Section titled “T-Cloud”T-Cloud is the label for models Telekom hosts. The catalog lists their region as Germany (EU). Enterprise Trust describes how prompts for these models are processed.
Partner models
Section titled “Partner models”Partner models are the models the API forwards to another provider. The catalog lists them on Microsoft Azure or Google Cloud, in a named region such as Sweden, France, or europe-, or worldwide. Enterprise Trust describes data handling for that path.
OpenAI-compatible
Section titled “OpenAI-compatible”The API works with the OpenAI SDK. The quickstart sets OPENAI_ and sends a request to chat/. Not every OpenAI parameter exists on every model, so use the guide for the call you are making.
Reasoning effort
Section titled “Reasoning effort”reasoning_ sets how much reasoning a model does before it answers. The datasheet lists the accepted values, and the list is different for each model. See Reasoning.
Embeddings
Section titled “Embeddings”An embedding is a vector for a piece of text, returned by POST /embeddings. Use embeddings for semantic search and retrieval. See Embeddings.