DEEPTHROAT DEVELOPER GUIDE
Reasoning controls
Choose a supported reasoning level for your latency and task requirements.
Available levels
Look up the selected model in GET /v1/models and use one of its efforts values. Supported levels differ between models.
Higher reasoning levels can take longer to produce the first token. Use a lower supported level for quick conversational tasks.
Set reasoning effort
{
"model": "gpt-6.1-sol",
"messages": [
{
"role": "user",
"content": "Review this function."
}
],
"reasoning_effort": "low",
"stream": true
}Omit reasoning_effort to use the model’s default behavior. For models with effort-based billing, the catalog shows the associated price multipliers.
Read reasoning output
When reasoning text is supplied, OpenAI-format responses expose message.reasoning_content or delta.reasoning_content during streaming. Anthropic-format responses can include thinking blocks. Some models return only the final answer.