Use OpenAI reasoning models on the Responses API: tune reasoning effort, request summaries, account for reasoning tokens, and carry reasoning items across turns including encrypted content

Documented steps

1. Set effort with reasoning:{effort:'medium'} on the Responses call. Supported values are model-dependent and can include none, minimal, low, medium, high, xhigh, and max.
2. Choose effort by workload: none or low for latency-sensitive work such as voice, retrieval, classification, and drafting; medium as the balanced default; high, xhigh, or max for deep debugging, long-horizon research, and complex agentic loops.
3. Check the specific model page before deploying, because a given model supports only a subset of effort values.
4. To get a reasoning summary, set summary:'auto' | 'concise' | 'detailed' in the reasoning object. 'auto' resolves to the most detailed summarizer that model supports.
5. Read the summary from the 'summary' array inside the reasoning output item. It is absent unless you explicitly opt in.
6. Measure cost at usage.output_tokens_details.reasoning_tokens. Reasoning tokens are billed as output tokens and consume context window.
7. Size max_output_tokens to leave headroom for reasoning tokens plus the visible answer, or generation can truncate.
8. On multi-step tool calling, pass the reasoning items returned with the last function call back on the next turn, alongside your function outputs, so the model continues its chain instead of re-deriving it.
9. Do this either via previous_response_id or by manually including all prior output items (reasoning, function_call, function_call_output) in the next request's input.
10. In stateless mode (store:false, or an organization on Zero Data Retention), reasoning items include an 'encrypted_content' property by default. Capture it and resend it on the next call in place of server-side state.
11. Note the legacy include value 'reasoning.encrypted_content' is still accepted for backward compatibility but is no longer required to receive encrypted_content.