Mastering Response Parameters

Working with the OpenAI Responses API

James Chapman

AI Curriculum Manager, DataCamp

Model Selection

response = client.responses.create(
    model="gpt-5-mini",
    input="What is recursion?",
)

¹ https://platform.openai.com/docs/models

Model Selection

response = client.responses.create(
    model="gpt-5-mini",
    input="What is recursion?",
)

Rule-of-thumb: use the cheapest model that meets the other requirements

Model	Speed	Cost	Best For
gpt-5-nano	Rapid	Ultra-Low	Low-latency apps
gpt-5-mini	Fast	Low	Simple tasks, Q&A
gpt-5	Moderate	Higher	Complex reasoning

¹ https://platform.openai.com/docs/models

LLMs and Tokens

Tokens: units of text that help the AI understand and interpret text

The sentence, "How can the OpenAI API deliver business value?" with each token highlighted in a different color.

¹ https://platform.openai.com/tokenizer

LLMs and Tokens

Non-reasoning LLMs: Generates the tokens with the highest probability of following the prompt

LLMs and Tokens

Reasoning Effort

response = client.responses.create(
    model="gpt-5-mini",
    input="Explain LLMs to a 6yr old.",

    reasoning={"effort": "minimal"}

)

Reasoning Effort

response = client.responses.create(
    model="gpt-5-mini",
    input="Explain LLMs to a 6yr old.",
    reasoning={"effort": "minimal"}
)

Effort	Best For
minimal	Trivial or mechanical tasks.
low	Simple tasks prioritizing speed and cost.
medium	Default for balanced reasoning and efficiency.
high	Complex, multi-step, or logic-heavy tasks.

Reasoning Summaries

response = client.responses.create(
    model="gpt-5-mini",
    input="Explain LLMs to a 6yr old.",
    reasoning={
        "effort": "medium",

        "summary": "auto"

    }
)

Limiting Output Tokens

response = client.responses.create(
    model="gpt-5-mini",
    input="Explain LLMs to a 6yr old.",
    reasoning={"effort": "minimal"},

    max_output_tokens=500

)

max_output_tokens includes the reasoning tokens

Summary

Simpler Tasks → start with small models, minimal reasoning, and few tokens
Complex Tasks → start with medium model sizes and reasoning, then fine-tune

Let's practice!

Working with the OpenAI Responses API