A practical reference for OpenAI-style clients using Jatevo model routes. This page describes request behavior, not benchmark quality, vendor claims, or output ranking.
Use model, messages, and stream for the most portable Chat Completions behavior across Jatevo routes.
Sampling fields may be accepted for SDK compatibility without guaranteeing precise output control.
Seeded generation, log probabilities, and multiple choices are not currently part of the GPT-5.5 surface.
Jatevo aims to be OpenAI-compatible for common Chat Completions clients. Some optional parameters from the wider OpenAI ecosystem are not available on every model route yet.
For production apps, prefer the minimal request shape unless a parameter is listed as supported for the exact model route you are using.
modelRequired. Use a model id available to your key, such as gpt-5.5 when enabled.
Call /v1/models with your key when you need to confirm available model ids.
messagesRequired for /v1/chat/completions. System, user, and assistant messages are accepted.
Keep system instructions stable and put task-specific context in user messages.
streamStreaming responses are supported for compatible clients.
Use stream: true for agent and CLI clients that need incremental output.
temperatureAccepted for client compatibility. On some model routes it may be normalized and may not change output.
Do not rely on temperature as a precise sampling guarantee for GPT-5.5 today.
top_pAccepted for client compatibility. On some model routes it may be normalized and may not change output.
Prefer prompt-level control for tone, length, and strictness.
nOnly n = 1 is supported. Requests for multiple choices are rejected.
Send separate requests if your application needs multiple independent candidates.
max_tokens / max_completion_tokensAccepted by many SDKs, but output length enforcement is not guaranteed on every GPT-5.5 route.
For now, include explicit length requirements in the prompt and validate output in your app.
seedDeterministic seed control is not currently available for GPT-5.5.
Avoid building workflows that depend on exact repeatability from the same seed.
logprobs / top_logprobsToken log probability output is not currently exposed on this model surface.
Use application-level scoring or a model route that explicitly documents logprob support.
presence_penalty / frequency_penaltyPenalty controls are not currently guaranteed for GPT-5.5 requests.
Use prompt instructions when you need less repetition or more variety.
logit_biasToken biasing is not currently supported on this model surface.
Do not depend on forced token inclusion or exclusion for safety-critical behavior.
stopStop sequences are not currently guaranteed for GPT-5.5 requests.
Post-process delimiters in your application when strict truncation is required.
This request shape is the safest baseline for OpenAI-compatible SDKs, CLI agents, and server-side integrations. Add optional controls only after confirming they are supported by the selected model route.
curl https://api.jatevo.ai/v1/chat/completions \
-H "Authorization: Bearer $JATEVO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5.5",
"messages": [
{
"role": "user",
"content": "Summarize this in three concise bullets."
}
],
"stream": true
}'This page reflects the current GPT-5.5 Chat Completions behavior on Jatevo. Parameter support can expand over time; applications should handle 400 responses for unsupported optional fields.