Parameters

The gateway accepts the standard OpenAI chat-completions parameters and passes them through to the upstream model. Support for individual parameters varies by model; unsupported parameters are ignored rather than rejected.

ParameterBehavior
modelModel slug, e.g. anthropic/claude-sonnet-5. Required.
messagesChat messages array with role and content. Required for chat completions.
max_tokensUpper bound on generated tokens.
temperatureSampling temperature; lower is more deterministic.
top_pNucleus sampling probability mass.
top_kTop-k sampling, where the model supports it.
frequency_penaltyPenalizes tokens by how often they have appeared.
presence_penaltyPenalizes tokens that have appeared at all.
repetition_penaltyMultiplicative repetition penalty, where supported.
stopOne or more sequences that halt generation.
streamStream tokens as server-sent events when true.
tools / tool_choiceFunction-calling definitions and selection behavior.
response_formatSet { "type": "json_object" } for JSON-mode output, where supported.
structured_outputsSchema-constrained output, where supported.
reasoningReasoning effort control for reasoning models: effort levels from minimal to xhigh.
seedBest-effort deterministic sampling seed.