Request Body

Request Body Parameters

Reference for the /v1/chat/completions POST body.

ParameterTypeRequiredDefaultDescription
modelstringYES-Model ID. Call GET /v1/models for available options.
messagesobject[]YES-Conversation messages. Each has role (system/user/assistant/tool), content (string), optional name (a-z, A-Z, 0-9, dashes, underscores, max 64).
thinkingobject|nullNonullObject with type: "enabled" or "disabled". At UMVA, default is disabled.
reasoning_effortstringNo"high"Values: "high", "max". "low"/"medium" map to "high", "xhigh" maps to "max".
max_tokensint|nullNomodel defaultMax tokens to generate.
response_formatobject|nullNo{"type":"text"}Output format: "text" or "json_object". For JSON, instruct model via system/user message.
stopstring|string[]|nullNonullUp to 16 stop sequences.
streambool|nullNofalseIf true, SSE streaming. Terminated by data: [DONE].
stream_optionsobject|nullNonullUse {"include_usage":true} for token usage before [DONE].
temperaturenumber|nullNo1Sampling 0-2. Higher=random, lower=deterministic. Alter this or top_p.
top_pnumber|nullNo1Nucleus sampling. 0.1 = top 10% mass. Alter this or temperature.
toolsobject[]|nullNonullFunction calling tools. Only "function" type. Each has name, description, parameters (JSON Schema). Max 128.
tool_choicestring|object|nullNo"none"/"auto""none" (no tools), "auto" (model decides), "required" (must call), or object to force a specific tool.
logprobsbool|nullNofalseReturn log probabilities of output tokens.
top_logprobsint|nullNonull0-20. Most likely tokens per position. Requires logprobs: true.
user_idstringNonullCustom identifier. Chars: [a-zA-Z0-9\-_], max 512. For content safety, KVCache isolation.

Deprecated Parameters

frequency_penalty and presence_penalty are no longer supported. Passing them has no effect.