OpenAI Chat

POST /v1/chat/completions

Example

{
  "model": "gpt-4o-mini",
  "messages": [
    { "role": "system", "content": "You are a helpful assistant." },
    { "role": "user", "content": "Describe Cubicspaces." }
  ],
  "service_tier": "default"
}

Processing Tier

For supported models, pass service_tier as a top-level request field, alongside model, messages, and stream.

{
  "model": "gpt-6-astra",
  "messages": [
    { "role": "user", "content": "Summarize this content quickly." }
  ],
  "service_tier": "fast"
}

Supported values:

Value	Description
default	Standard processing for regular production requests
flex	Flexible processing for lower-priority, asynchronous, or batch-style work. Responses may be slower and can be unavailable when capacity is constrained
fast	Faster processing for latency-sensitive requests
priority	Compatibility alias for fast processing
auto	Use the server or project default tier

The response service_tier shows the tier that was actually used for the request. It may differ from the value sent by the client.

GPT-6 Astra Compatibility

When calling gpt-6-astra:

• Use max_completion_tokens instead of max_tokens.
• reasoning_effort accepts low, medium, high, and xhigh.
• Do not send temperature, top_p, top_k, frequency_penalty, presence_penalty, logprobs, stop, or chat_template_kwargs.
• New applications that combine reasoning with tool calls should use /v1/responses directly.
• The legacy /v1/completions endpoint does not support this model.
• Cache writes are reported as usage.prompt_tokens_details.cache_write_tokens.

Compatibility mode may remove unsupported sampling fields and translate max_tokens to max_completion_tokens. Clients should still send the model's native parameters for consistent behavior in every request.